add describe-time producer and override-route hints

[HumanBean17]

Summary

Closes #161.

• Add five v1 describe hint templates in mcp_hints.py for the producer axis (DECLARES.DECLARES_PRODUCER, DECLARES_PRODUCER, OVERRIDDEN_BY.DECLARES_PRODUCER) and the missing override-route axis (OVERRIDDEN_BY.EXPOSES), plus producer-node declaring-method hint.
• Wire triggers in generate_hints with the same priorities as symmetric client/route hints.
• Add integration and synthetic tests in tests/test_mcp_hints.py.

No schema change, no re-index, no ontology bump.

Test plan

• .venv/bin/ruff check .
• .venv/bin/python -m pytest tests/test_mcp_hints.py -v (110 passed, 2 skipped)
• .venv/bin/python -m pytest tests -v (579 passed; 18 CLI failures are pre-existing env issue — system java-codebase-rag not editable venv install)

Made with Cursor

[HumanBean17]

Review summary

Verdict: Approve — small, correctly scoped fix for #161. I don't see a functional bug in the hint logic.

What works well

• All five missing describe hints are implemented with the expected traversal shapes, atomic EdgeType literals (no dot-keys in emissions), and the same priority bands as existing client/route hints.
• Producer leaf nodes follow the same early-return pattern as route / client; type and method branches mirror the symmetric triggers.
• Tests are mostly at the right layer: bank-chat e2e for type/method/producer leaf hints; override_axis_rollup_smoke e2e for OVERRIDDEN_BY.EXPOSES; parametrized ≤120-char sweep for all new templates.
• No unnecessary schema/ontology/re-index churn — accurate for this PR.

Nits (non-blocking)

1. OVERRIDDEN_BY.DECLARES_PRODUCER coverage is weakest — test_hints_describe_method_producers_in_overriders_emits is synthetic generate_hints(...) only, while clients_in_overriders uses bank-chat and routes_in_overriders uses override_axis_graph + real describe_v2. No fixture currently produces that rollup key. Follow-up: extend override_axis_rollup_smoke (or a tiny brownfield stub) with override methods that declare producers, mirroring test_describe_abstract_method_with_route_override_emits_exposes.

2. Propose/catalog hygiene — mcp_hints.py still references the locked v1 catalog in HINTS-ROAD-SIGNS-PROPOSE.md Appendix A, which doesn't list these five templates. Worth a short amendment note so the lock doc and code don't diverge.

3. docs/AGENT-GUIDE.md drift — override-axis section still documents only OVERRIDDEN_BY.DECLARES_CLIENT / OVERRIDDEN_BY.EXPOSES, not OVERRIDDEN_BY.DECLARES_PRODUCER (already in mcp_v2.py / server.py). Optional doc follow-up.

4. Tiny test inconsistency — parametrized length test uses {"id": "a"} for TPL_DESCRIBE_METHOD_PRODUCERS_IN_OVERRIDERS while siblings use sym:a (harmless for length-only).

5. Duplicated override_axis_graph fixture — copied from test_mcp_v2_compose.py; fine for this PR, could move to conftest.py later.

Note on hint cap

A method with full override-axis rollups plus multiple leaf edges can exceed five candidates; finalize_hint_list drops lower-priority leaf hints first. Pre-existing behavior, not introduced here.

Please confirm CI green on this branch (Actions was pending at review time).