chore(terminology): adopt agreed Granite Switch / Mellea glossary (#1192)

[planetf1]

The Epic #929 team agreed a canonical glossary for Granite Switch / Mellea naming, but most existing prose still used the old vocabulary — "intrinsic", "activated-LoRA", "Project Granite Switch". This PR sweeps the Phase 0 code files, AGENTS.md, and the reference glossary to use the agreed terms consistently, before Phase 1 contributors carry old vocabulary into new work.

Python symbol names (IntrinsicAdapter, fetch_intrinsic_metadata, etc.) and import paths are untouched — those renames are deliberately deferred to Phase 1 (#1136) and Phase 4 (#1144). The prose/symbol mismatch that results is intentional and documented in the glossary's Intrinsic entry.

Where this fits

Epic #929 — adapter lifecycle. Sits before Phase 1 (#1136 symbol renames) and addresses the "Done when" items of #1192 directly.

Fixes #1192

Changes

mellea/backends/adapters/ (4 files) + mellea/stdlib/components/adapter_based_component/__init__.py

• Comments and docstrings only; no symbol renames
• "intrinsic function" → "adapter function" in all free prose
• "Activated LoRA" → "aLoRA" in the AdapterType.ALORA enum doc
• catalog.py Field(description=...) strings updated (user-visible prose)

AGENTS.md

• Section ## 13. Working with Intrinsics → ## 14. Working with Adapter Functions
• Fixes a duplicate ## 13 heading (was two sections both numbered 13)
• Import paths, Intrinsic class name, docs/examples/intrinsics/ path unchanged

docs/docs/reference/glossary.md

• Rewrites the aLoRA entry with the agreed KV-cache-sharing definition (cache hits, not "cache rewrite")
• Rewrites the Granite Switch entry with the agreed architecture definition
• Updates the Intrinsic entry to point to Adapter function as canonical, noting Intrinsic is the current Python name pending Phase 1/4
• Adds 8 new entries: Adapter function, Adapter, Capability, Checkpoint, Granite Libraries (with Core/RAG/Guardian table), Granite Switch composer, Mellea, and a References section

Testing

• uv run ruff format + uv run ruff check — clean
• uv run pytest test/ -m "not qualitative" — 2083 passed, 7 pre-existing failures (missing huggingface_hub extra — unrelated to this PR, confirmed by stash test)
• markdownlint (docs) pre-commit hook — passed

[ajbozarth]

LGTM on the executable-code side — pure prose changes, zero risk surface, and the new glossary entries read well. Two things before merge:

1. The terminology sweep missed a number of prose hits across the repo. Some are inside files this PR already touches (worth fixing here); others are scattered across backends, docstrings, and integration docs (Phase 0 follow-up — your call whether to fold in or defer).
2. Alphabetical ordering nit on the new Adapter / Adapter function entries — left as a separate inline suggestion.

Note: The replacements below were generated by an AI assistant scanning for the old vocabulary; they reflect the conventions established by this PR's own diff (aLoRA adapter(s) for the plural, adapter function for prose) but may not be 100% correct in every case. Treat the table as a checklist to review, not a mechanical patch — sentence flow may need tweaks beyond the literal find/replace, and a couple of judgment calls are flagged at the bottom.

Remaining old vocabulary

In files this PR already modifies (recommend fixing here)

File	Line	Find	Replace with
docs/docs/reference/glossary.md	382	A Guardian Intrinsic function that	A Guardian adapter function that
docs/docs/reference/glossary.md	398	A Guardian Intrinsic function that	A Guardian adapter function that
docs/docs/reference/glossary.md	412	A Guardian Intrinsic function that	A Guardian adapter function that
docs/docs/reference/glossary.md	458	A Guardian Intrinsic function that	A Guardian adapter function that

Out of scope for this PR (file a follow-up issue, or fold in if you'd rather sweep once)

Library code — module/function docstrings:

File	Line	Find	Replace with
mellea/stdlib/components/intrinsic/core.py	1	Intrinsic functions for core model capabilities.	Adapter functions for core model capabilities.
mellea/stdlib/components/intrinsic/core.py	15	Intrinsic function that evaluates	Adapter function that evaluates
mellea/stdlib/components/intrinsic/core.py	35	Intrinsic function that determines	Adapter function that determines
mellea/stdlib/components/intrinsic/core.py	62	Intrinsic function that finds	Adapter function that finds
mellea/stdlib/components/intrinsic/rag.py	1	Intrinsic functions related to retrieval-augmented generation.	Adapter functions related to retrieval-augmented generation.
mellea/stdlib/components/intrinsic/rag.py	21	Intrinsic function that checks whether the question	Adapter function that checks whether the question
mellea/stdlib/components/intrinsic/rag.py	54	Intrinsic function that rewrites	Adapter function that rewrites
mellea/stdlib/components/intrinsic/rag.py	82	Intrinsic function that determines	Adapter function that determines
mellea/stdlib/components/intrinsic/rag.py	119	Intrinsic function that finds	Adapter function that finds
mellea/stdlib/components/intrinsic/rag.py	165	Intrinsic function that checks whether a single document	Adapter function that checks whether a single document
mellea/stdlib/components/intrinsic/rag.py	204	Intrinsic function that checks whether the sentences	Adapter function that checks whether the sentences
mellea/stdlib/components/intrinsic/guardian.py	1	Intrinsic functions for Guardian safety and hallucination detection.	Adapter functions for Guardian safety and hallucination detection.
mellea/stdlib/components/intrinsic/guardian.py	246	Intrinsic function that evaluates the factuality	Adapter function that evaluates the factuality
mellea/stdlib/components/intrinsic/guardian.py	262	Intrinsic function that corrects	Adapter function that corrects

Library code — backend "Activated LoRA" / "ALora" prose:

File	Line	Find	Replace with
mellea/backends/huggingface.py	3	Span-based context or ALoras	Span-based context or aLoRA adapters
mellea/backends/huggingface.py	242	This backend also supports Activated LoRAs (ALoras)](https://arxiv.org/pdf/2504.12397).	This backend also supports [aLoRA adapters](https://arxiv.org/pdf/2504.12397).
mellea/backends/litellm.py	318	The LiteLLM backend does not support activated LoRAs.	The LiteLLM backend does not support aLoRA adapters.
mellea/backends/watsonx.py	378	The watsonx backend does not support currently support activated LoRAs.	The watsonx backend does not currently support aLoRA adapters.
mellea/backends/ollama.py	392	The ollama backend does not support currently support activated LoRAs.	The ollama backend does not currently support aLoRA adapters.
mellea/core/requirement.py	263	This includes ALora because	This includes aLoRA because

Docs:

File	Line	Find	Replace with
docs/docs/integrations/openai.md	264	these adapters enable intrinsic functions	these adapters enable adapter functions
docs/docs/integrations/openai.md	293	The high-level intrinsic wrappers	The high-level adapter function wrappers
docs/docs/integrations/openai.md	295	for the full list of available intrinsics.	for the full list of available adapter functions.
docs/docs/integrations/openai.md	299	Only intrinsics	Only adapter functions
docs/docs/integrations/huggingface.md	86	[Activated LoRA (aLoRA)]	[aLoRA]
docs/docs/integrations/huggingface.md	90	For intrinsics without local GPU requirements	For adapter functions without local GPU requirements
docs/docs/integrations/huggingface.md	92	[Intrinsics](../advanced/intrinsics)	[Adapter Functions](../advanced/intrinsics.md)
docs/docs/advanced/lora-and-alora-adapters.md	21	pre-trained intrinsic adapters embedded	pre-trained adapter functions embedded
docs/docs/advanced/lora-and-alora-adapters.md	23	See [Intrinsics](./intrinsics) for details.	See [Adapter Functions](./intrinsics.md) for details.
docs/docs/advanced/lora-and-alora-adapters.md	107	as a Mellea intrinsic (so that it can	as a Mellea adapter function (so that it can
docs/docs/advanced/lora-and-alora-adapters.md	163	[Intrinsics](./intrinsics)	[Adapter Functions](./intrinsics.md)
docs/docs/advanced/prefix-caching-and-kv-blocks.md	135	[Intrinsics](./intrinsics)	[Adapter Functions](./intrinsics.md)

Intentionally not flagged

• The --intrinsic CLI flag and mellea/stdlib/components/intrinsic/ import paths in docs/docs/advanced/lora-and-alora-adapters.md and AGENTS.md — these are symbol/path names, deferred to Phase 1 per the PR description.
• docs/docs/api/mellea/backends/adapters/adapter.mdx — autogenerated; will refresh from source on the next docs build.
• glossary.md:54 — aLoRA (Activated LoRA) is the entry's own definitional expansion, not stale prose.

Judgment calls flagged for review

• The watsonx.py:378 and ollama.py:392 strings contain a duplicated "does not support currently support" — strict scope is terminology only, but fixing the duplication while you're touching the line is cheap. Drop the typo fix from the replacement if you'd rather keep the PR minimal.
• lora-and-alora-adapters.md and huggingface.md doc cross-links currently lack the .md suffix that docs/CONTRIBUTING_DOCS.md requires. Adding it where we're already touching the line is consistent; ignore those columns if you'd rather keep the link-suffix sweep separate.

Verification command

After applying the table, this should return zero hits within prose (excluding paths/symbols):

grep -rnE "intrinsic function|Intrinsic function|Activated LoRA|activated[- ]LoRA|\bALoras?\b" \
  mellea/ docs/docs/ \
  | grep -v "/test/" \
  | grep -v "\.mdx:" \
  | grep -v "stdlib\.components\.intrinsic\|formatters\.granite\.intrinsics"

[planetf1]

Still looking at some related usages around guardian

[ajbozarth]

All my review items have been addressed, LGTM

[planetf1]

Remaining before merge

1. "Guardian adapter functions" — collective term needs confirmation

~12 occurrences of "Guardian Intrinsics" used as a group label remain unchanged in: `safety-guardrails.md`, `common-errors.md`, `metrics.md`, `concepts/architecture-vs-agents.md`, `examples/index.md`, `use-context-and-sessions.md`, `stdlib/requirements/safety/guardian.py`.

Not changed: the glossary defines per-function descriptions ("A Guardian adapter function") but does not give an agreed collective replacement. Options: "Guardian adapter functions", "Guardian Library", or keep "Guardian Intrinsics" as a proper-noun label.

2. Four cross-doc link anchors (may fit in followon issue)

All point to the still-titled "Intrinsics" page; deferred to Phase 1 page rename.

File	Current	Alex suggests
`docs/docs/integrations/huggingface.md:92`	`Intrinsics`	`Adapter Functions`
`docs/docs/advanced/lora-and-alora-adapters.md:23`	`Intrinsics`	`Adapter Functions`
`docs/docs/advanced/lora-and-alora-adapters.md:163`	`Intrinsics`	`Adapter Functions`
`docs/docs/advanced/prefix-caching-and-kv-blocks.md:135`	`Intrinsics`	`Adapter Functions`

Continuing next week & look for any additional omissions