fix(rag): align retrieval quality knobs with reviewbot — minScore 0.4 + bm25 rerank + title-led query (#GAP-2)

[JSONbored]

Summary

Aligns the gittensory codebase-RAG retrieval seam (src/review/rag-wire.ts) with reviewbot's core config, fixing GAP-2 (the code half).

Before, the retrieve call used minScore: 0 (off) and no reranker, so low-relevance cosine "neighbours" were injected as "RELEVANT EXISTING CODE" — noise that itself drives false positives. The query also omitted the PR title, hurting recall vs. reviewbot (which embeds ${title}\n${diff}).

Changes

• src/review/rag-wire.ts
  • buildRagQuery(files, title?) — PREPENDS the (trimmed) PR title to the query text; additive, omitted cleanly when blank/absent.
  • buildReviewRagContext(...) — always passes minScore: 0.4 + reranker: "bm25" to retrieveContext (matches reviewbot rag: { minScore: 0.4, reranker: "bm25" }). A caller-supplied reranker still wins (e.g. forcing "off"); minScore is fixed.
• src/queue/processors.ts — feeds the PR title into the retrieval-context build at the (already flag+allowlist-gated) review call site.
• test/unit/rag-wiring.test.ts — new assertions: title leads the query; retrieveContext invoked with minScore: 0.4 / reranker: "bm25"; caller override honored; blank-title handling.

Scope / safety

Code knobs ONLY. RAG stays per-repo opt-in (isRagEnabled(env) flag AND the convergence repo allowlist) — this does not force-enable RAG globally; the operator enables + warms the index via runtime config separately. Fully fail-safe path unchanged.

Note: Task A (repoint public-stats to the live audit_events ledger) was already shipped on main in #1114, so this branch contains only the RAG (Task B) work.

Verification

• npx tsc --noEmit — clean (exactOptionalPropertyTypes + noUncheckedIndexedAccess)
• npx vitest run — 3546 passed, 1 skipped (215 files); rag-wiring suite 16 → 21 tests

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.84%. Comparing base (6be41be) to head (292cbe4).
⚠️ Report is 1 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #1124   +/-   ##
=======================================
  Coverage   94.83%   94.84%           
=======================================
  Files         155      155           
  Lines       18708    18711    +3     
  Branches     6772     6773    +1     
=======================================
+ Hits        17742    17746    +4     
  Misses        395      395           
+ Partials      571      570    -1     

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

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.