Honor rate-limit Retry-After across Google/Anthropic/GitHub SDKs

[hiroshinishio]

Summary

When the Gemini free-tier quota trips (429 RESOURCE_EXHAUSTED), the API returns a "Please retry in N.NNNs" hint in the error message body. Previously chat_with_google just raised, cascaded through chat_with_model → chat_with_agent → handle_webhook_event, and every Lambda in the incident window died on Sentry (AGENT-3K5/3K6/3K7/3K8/36M/36Q, the gitautoai/website run on 2026-04-20 16:23 UTC).

This PR generalizes rate-limit handling across every SDK we talk to.

• New utils/error/get_rate_limit_retry_after.py dispatches per error type:
  • requests.HTTPError → GitHub headers (X-RateLimit-Remaining=0 + X-RateLimit-Reset primary, Retry-After secondary) via parse_github_rate_limit_headers, or generic Retry-After via parse_retry_after_header
  • Anthropic APIStatusError (status_code=429) → retry-after header
  • Google ClientError (code=429) → "Please retry in N.NNNs" in message body via parse_google_retry_in_message
• handle_exceptions calls it from both the requests.HTTPError and generic Exception branches. If a delay comes back AND the existing TRANSIENT_MAX_ATTEMPTS budget allows, it sleeps the honored delay and continues the retry loop. No upper cap — should_bail at the handler layer already enforces Lambda timeout.
• The old handle_github_rate_limit and the api_type=="github" special case in handle_http_error are gone. Both are now just input shapes for the same generic extractor.
• Test fixture (fixtures/real_google_429.txt) is the verbatim CloudWatch log line from the gitautoai/website incident (Please retry in 59.739387544s), preserved with full details[] payload.

Social Media Post (GitAuto)

Rate-limit 429s from Google/Anthropic/GitHub now retry cleanly instead of crashing the Lambda

• New helper pulls the retry-after delay from any SDK's error shape (Google's message body, Anthropic's retry-after header, GitHub's X-RateLimit-Reset)
• handle_exceptions sleeps the honored delay and retries under the existing transient-retry budget
• Old GitHub-specific rate-limit path folded into the generic handler, one code path for every SDK

Social Media Post (Wes)

The Gemini free-tier 429 takes a different shape than GitHub's 429, which takes a different shape than Anthropic's 429, and I had a per-SDK handler for exactly zero of them. Wrote one extractor that dispatches on error type, hooked it into the existing retry loop, honored whatever delay the server suggested. Three error shapes, one path, no more 429s in Sentry.