expand bank-chat fixture with brownfield integration coverage

[HumanBean17]

Summary

• Enriches tests/bank-chat-system/ with notification, audit, reporting, HTTP/async integration, and in-corpus @Codebase* brownfield annotations (definitions in chat-contracts, usage across assign/app/engine).
• Adds Tier-1 guardrails in tests/test_bank_chat_brownfield_integration.py for brownfield routes, clients, async Layer-C, and @CodebaseProducers on the session bank graph (pass1–5, no pass6).
• Fixes tests exposed by the larger corpus: test_find_client_by_client_kind (verify client_kind via graph rows, not display fqn) and test_increment_updates_lance_after_touch_java_file (touch file under Maven src/main/java).
• Updates tests/README.md fixture tree and Feign/brownfield guidance.

Test plan

• .venv/bin/ruff check .
• .venv/bin/python -m pytest tests -q — 698 passed, 8 skipped
• Graph meta on rebuilt bank corpus: routes_from_brownfield_pct ~33%, http_clients_from_brownfield_pct ~12.5%, async_producers_from_brownfield_pct ~33%

No indexer / ontology_version changes (fixture + tests only).

Made with Cursor

[HumanBean17]

Review summary

Direction: Good — fills a real Tier-1 gap that tests/fixtures/brownfield_* stubs cannot cover (annotations defined in chat-contracts, consumed across assign / app / engine on the session bank graph).

Necessity: Partially. The brownfield integration story, Feign client_kind coverage, CLI increment touch-path fix, and test_bank_chat_brownfield_integration.py guardrails are warranted. Most of the ~1.3k LOC of notification/audit/reporting domain is not exercised by any new assertion — only 8 @Codebase* usages drive the new tests. A ~10–15 file focused slice would satisfy the same guardrails and align better with tests/README.md rule 5 (“keep the fixture small”).

What works well

• Tier-1 layout matches README: contracts hold @Codebase* definitions; usage spans modules; session graph pass1–5 (no pass6).
• test_find_client_by_client_kind fix is correct — display fqn does not embed client_kind; verify via list_clients() / graph columns after Feign clients were added.
• test_increment_updates_lance_after_touch_java_file — touch under Maven src/main/java matches indexer walk semantics.
• No production/indexer/ontology_version churn; assertions stay loose (>= 1, set checks, meta > 0).

Suggestions (non-blocking)

1. Fixture size: Consider trimming notification/audit/reporting bulk unless a follow-up test or manual-verification checklist needs it — every Tier-1 test pays session graph build cost on the larger tree.
2. PR hygiene: A short propose (Tier-1 goals, bank-chat vs brownfield_* split, timing before/after) would match repo culture for a ~1.4k-line fixture change; README also asks for wall-time notes on large refactors.
3. Minor: ChatCoreFeignClient importing JoinOperatorRequest from chat-app is a slightly odd cross-module edge in a toy corpus — fine if intentional.

Recommendation

Approve the intent (Tier-1 brownfield + test fixes). Optional follow-up: trim unused domain files or document why they stay (e.g. future describe/hints scenarios). Not a reason to reject — the gap and fixes are real.