feat(sleep): group mined tasks by skill hint with catch-all fallback

[bogdanbaciu21]

Summary

Third review-sized slice of #120: the grouping primitive, so mined tasks can be
routed per skill while everything without clear evidence stays in the existing
managed-skill catch-all.

Adds mine.group_tasks_by_skill_hint(tasks, managed_skill_name), a
deterministic, side-effect-free helper returning ordinary insertion-ordered
dicts/lists:

• groups preserve first-seen order, and so does task order inside a group;
• each task ID is emitted exactly once, reusing dedup_tasks merge semantics
  (source sessions unioned, best outcome kept);
• a task reaches a hinted group only when every observation of that ID agrees
  on the same non-empty hint;
• absent/blank hints, two different hints for one ID, and a mixture of hinted and
  unhinted observations for one ID all route to managed_skill_name;
• a hint equal to managed_skill_name lands in that one group, never a duplicate.

Nothing else changes: mine(), filter_tasks_for_target(), cycle, gates,
staging, and adoption are untouched, and the caller supplies the configured
managed-skill name rather than the helper resolving paths or reading skill files.

Stacked on #182 and #183, which add SessionDigest.skills_used and
TaskRecord.skill_hint. Those commits are this branch's parents; reviewed in
order this diff is just the grouping commit. Happy to rebase/squash to taste.

Tests

• python -m pytest -q tests/test_sleep_engine.py -k group_tasks_by_skill_hint → 8 passed
• python -m py_compile skillopt_sleep/mine.py tests/test_sleep_engine.py → clean
• python -m pytest -q → 588 passed, 7 skipped (base main at 8304e6c:
  556 passed, 7 skipped)

Covered: empty input, legacy hint-less tasks, blank hint, interleaved
alpha/beta/alpha ordering, duplicate ID merge, conflicting hints, partial
hint evidence, and a hint equal to the managed skill name.

Refs #120

[Copilot]

Pull request overview

This PR adds the remaining “grouping primitive” needed to route mined sleep-engine tasks per skill, using harvested Skill tool invocations as evidence, while keeping ambiguous or hint-less tasks in the existing managed-skill catch-all.

Changes:

• Add SessionDigest.skills_used harvesting from Claude Skill tool-use blocks and propagate a derived TaskRecord.skill_hint into both heuristic and LLM miners.
• Introduce mine.group_tasks_by_skill_hint(tasks, managed_skill_name) to deterministically group (and deduplicate) tasks by unambiguous skill hint, with a managed-skill fallback.
• Add focused unit tests covering harvesting, hint propagation, dedup semantics, and grouping behavior.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
tests/test_sleep_skill_hint.py	Adds unit tests for skill_hint propagation, serialization, and dedup behavior.
tests/test_sleep_skill_harvest.py	Adds unit tests for harvesting Claude Skill tool invocations into skills_used.
tests/test_sleep_engine.py	Adds coverage for group_tasks_by_skill_hint behavior and ordering guarantees.
skillopt_sleep/types.py	Extends core dataclasses with SessionDigest.skills_used and TaskRecord.skill_hint.
skillopt_sleep/mine.py	Adds session_skill_hint, propagates hints in heuristic mining, extends dedup_tasks, and introduces group_tasks_by_skill_hint.
skillopt_sleep/llm_miner.py	Propagates session-derived skill_hint into LLM-mined tasks.
skillopt_sleep/harvest.py	Extracts Skill targets from tool-use blocks and persists them into digests.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[bogdanbaciu21]

Thanks for the welcome on #120. To keep review load bounded and follow the incremental shape Yif-Yang suggested, I'm closing this PR for now and will reopen it once #182 (the harvesting slice) lands or the maintainer asks for the next slice. The branch stays on my fork so this can be re-opened as-is. Happy to restructure if a different cadence is preferred.