Add a hardware-free naming API for static PV-name prediction

[gilesknap]

Adds fastcs_catio.naming — predict_chain() / predict_names() — which derive
the PV names an EtherCAT chain will get without a bus, a controller, or an
event loop.

No open issue covers this; it is driven by a downstream consumer,
epics-containers/builder2ibek#128.

Why

At runtime the controller names are a side effect of ADS discovery:
FastCSClient._get_ethercat_chains stamps each slave's ChainLocation, and
CATioServerController._resolve_controller_name_and_path renders it through
CATioNameMappings. Both need live hardware.

A site migrating an existing installation onto fastcs-catio needs those same
names before any hardware exists — other IOCs hold PV references that must be
rewritten ahead of the switchover. DLS is doing exactly this: replacing the
legacy ethercat scanner beamline by beamline, where the new names differ from
the old ones and every consumer IOC's database has to be migrated in step.

What it does, and what it deliberately does not

Purely additive. catio_controller.py and client.py are untouched; the
runtime path does not change. The diff is two new files.

Keyed on terminal type names ("EL3104"), not on
(vendor_id, product_code, revision_number). A static description of a chain is
all a migration tool has — DLS builder XML carries type_rev="EL3104 rev 0x00120000" and nothing else — whereas the runtime keys on the identity
reported over ADS. Looking up by type name is revision-independent, which
mirrors the vendor+product fallback get_terminal_type_by_identity already
applies when a rig runs newer firmware than the cached description.

strict=True raises UnknownTerminalTypeError for a terminal missing from
terminal_types.yaml. The runtime is deliberately forgiving here — an
unrecognised identity renders as MOD<position> — but that is the wrong default
for a migration tool, which would then rewrite live PV references to names the
IOC never creates. Callers generating names for real hardware should ask for the
error; strict=False reproduces the runtime's behaviour.

The part worth reviewing most

tests/test_naming.py is the anti-drift mechanism. 18 of its 27 tests drive the
real runtime code path — _get_ethercat_chains, _generate_system_tree,
_compute_module_alias_indices, _resolve_controller_name_and_path — over 6
chain shapes × 3 template shapes, and assert predict_names() agrees. A
reimplementation that silently drifts from the runtime is the obvious failure
mode for a module like this, so the tests are written to break in this repo,
where the change that caused the drift lives, rather than downstream.

One deliberate oddity: the helper uses a private event loop and restores the
previous one, rather than asyncio.run(). asyncio.run clears the
process-wide loop on exit, which strands the loop tests/test_system.py
installs for its ADS simulator — that loop and its sockets are then collected
unclosed and pytest raises PytestUnraisableExceptionWarning against an
unrelated test. It is invisible when the file runs alone and only shows up in a
full-suite run, so please don't simplify it back.

Base branch — please read before merging

This targets gui/inline-root-controllers (#63), not main, because
naming.py needs CATioNameMappings and get_terminal_type(...).group_alias,
and a tree-wide grep of main's src/ returns zero hits for either. Against
main it does not merely misbehave, it fails at the import line. Both arrive
with #63.

If #63 merges first, retarget this to main — it should need no other change.

Instructions to reviewer on how to test:

1. uv run pytest — expect 267 passed, 23 skipped (240 on the base branch,
   plus 27 new tests and 1 new doctest).
2. uv run pyright src tests — expect 0 errors.
3. git diff gui/inline-root-controllers --stat — confirm only two files are
   added and nothing existing is modified.
4. Read the predict_names doctest for the shape of the output:
   a (node, position) -> PV prefix mapping.

Checks for reviewer

• Would the PR title make sense to a user on a set of release notes

[codecov]

Codecov Report

❌ Patch coverage is 91.07143% with 10 lines in your changes missing coverage. Please review.
✅ Project coverage is 76.30%. Comparing base (a37e7d6) to head (fdfdd2b).

Files with missing lines	Patch %	Lines
src/fastcs_catio/naming.py	91.07%	10 Missing ⚠️

Additional details and impacted files

@@                       Coverage Diff                       @@
##           gui/inline-root-controllers      #65      +/-   ##
===============================================================
+ Coverage                        75.91%   76.30%   +0.39%     
===============================================================
  Files                               19       20       +1     
  Lines                             4197     4309     +112     
===============================================================
+ Hits                              3186     3288     +102     
- Misses                            1011     1021      +10     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.