fix: boost exact username search matches

[igorrendulic]

Summary

Support search now treats exact Telegram author usernames as a strong retrieval signal. Searching crinx7 or @crinx7 returns normal source-linked Telegram evidence authored by that user, instead of letting weaker text mentions or semantic matches outrank it.

The boost is based on author_username metadata through the shared Local Core, not raw username mentions inside message text. Matching evidence keeps the existing Evidence Bundle shape, so CLI search, draft context, and agent workflows keep consuming the same result contract.

Validation

• uv run pytest -> 120 passed
• uv run --with ruff ruff check . -> passed
• Multi-agent code review found one P2 performance issue; fixed by moving exact author lookup to the indexed storage path and rerunning validation.

Post-Deploy Monitoring & Validation

• Validation window: first support-search sessions after merge.
• Owner: support operator running local searches.
• Healthy signals: searches for known Telegram usernames such as crinx7 or @crinx7 return Telegram evidence with matching metadata.author; normal topic searches still return policy/manual/web evidence as before.
• Failure signals: username searches return raw mention-only results first, return standalone user records instead of evidence, or Manual Knowledge Note conflict behavior changes for non-username queries.
• Mitigation trigger: if normal support-topic search quality regresses, revert the username boost commit or temporarily remove the username candidate signal from hybrid fusion.

---