Handle transient GitHub API connection resets with retries

[cursor]

Summary

• add bounded retry logic to GithubClient._fetch_github() for transient GitHub transport failures (ConnectionError, Timeout)
• add an explicit request timeout for GitHub API calls to avoid hanging network operations
• preserve current behavior for non-transient HTTP failures by still raising HTTPError
• add tests covering retry-until-success and retry-exhaustion paths

Root cause

_fetch_github() used a single, no-timeout requests.get() call. A transient upstream/network reset (Connection reset by peer) caused the request to fail immediately and bubble up as a 500, producing issue SENTRY-GITHUB-ACTIONS-APP-71.

Validation

• python3 -m pytest -q passes
• new tests verify exponential backoff retries (1s, 2s) and proper failure on final exhaustion

Risk

• low risk: retry scope is limited to transient connection/timeout failures and capped at 3 attempts
• potential tradeoff: worst-case request latency increases on repeated transient failures due to bounded backoff

Fixes SENTRY-GITHUB-ACTIONS-APP-71