* Fix Lark recent context fetch degradation
* fix(lark): don't retry doomed recent-context fetches
Two retry-policy corrections on the recent-context degradation path:
- Skip the retry when the shared enrichment context is already done.
Both attempts reuse one ctx capped by EnrichTimeout (~2s), so a
first-attempt deadline can never recover -- retrying just burns a
doomed call. Guard the retry on ctx.Err().
- Stop treating rate limits (230020 / HTTP 429) as retryable. The client
drops Retry-After, so an immediate second call re-hits the limit and
doubles list load on an already-throttled tenant; degrade instead.
Tests: replace the misleading DeadlineExceeded "recovers on retry" case
with a transient network error (ctx still valid); add coverage for the
done-context no-retry guard, rate-limit no-retry, and the production
plain-wrapped-error classifier path (permission / deleted / token).
Co-authored-by: multica-agent <github@multica.ai>
---------
Co-authored-by: J <j@multica.ai>
Co-authored-by: multica-agent <github@multica.ai>
51e4afc480