- Add _ThreadContextCache dataclass for caching fetched context (60s TTL)
- Add exponential backoff retry for conversations.replies 429 rate limits
(Tier 3, ~50 req/min)
- Only fetch context when no active session exists (guard at call site)
to prevent duplication across turns
- Hoist bot_uid lookup outside the per-message loop
- Clearer header text for injected thread context
Based on PR #6162 by jarvisxyz, cherry-picked onto current main.
jarvisxyz
d12f8a157c