Aurora retry logging: accurate reasons, correct levels, never throw

[wparad]

Summary

• Fixed the aurora.retry WARN log, which previously said only "Aurora query failed — retrying" with no indication of why. It now includes the actual failure reason.
• The reason is read from the unwrapped SDK/driver cause rather than DrizzleQueryError's own .message, which dumps the raw SQL and bound parameter values (e.g. a thread ID) — that stays out of the log title/message while the full error (query text included) is still available via the error context field.
• Downgraded the per-attempt retry log to INFO specifically when the failure is Aurora resuming from auto-pause: expected and self-resolving, with nothing actionable — already flagged once by the existing aurora.resume_detected INFO log. Non-resume transient failures still log at WARN.
• withRetry() in src/database/thread-matcher.ts no longer throws — it returns Result<T, DbError>. Every terminal failure comes back as an err(DbError) carrying the operation name and the attempt count it gave up at (new optional operation/attempts fields on DbError in src/errors.ts), so callers get full context without wrapping it in their own try/catch.
• Every ThreadMatcher method now checks that Result directly instead of wrapping withRetry in try/catch and re-deriving a DbError from a caught exception.
• processor.ts's tier-2 similarity search previously discarded ThreadMatcher.findMatch() errors with no logging at all; it now logs a WARN (matching the existing tier-1.5 In-Reply-To pattern) before degrading to "no match".

Test plan

• npx tsc --noEmit passes
• npx eslint src/database/thread-matcher.ts src/errors.ts src/processor/processor.ts passes clean
• Full test suite: npx vitest run — 2846/2846 pass

---

Generated by Claude Code