Wrap GitHub API calls in a retrying client with exponential backoff

[Copilot]

GitHub API interactions were executed directly from the service layer, so transient failures could cause issue sync operations to fail immediately. This change introduces a thin wrapper around the existing go-github client that preserves current call shapes while applying retry/backoff to retryable failures.

• Retrying GitHub API wrapper

  • Added a dedicated githubAPI client that mirrors the GitHub operations used by the action.
  • Encapsulated retry behavior in a shared loop, with per-operation doXYZ methods invoked inside the retry path.

• Backoff and retry policy

  • Uses github.com/jpillora/backoff for exponential backoff with jitter.
  • Retries transient GitHub failures only: rate limiting, secondary rate limiting, accepted/pending responses, HTTP 429, HTTP 5xx, and timeout-based network errors.
  • Preserves context cancellation while waiting between attempts.

• Service wiring

  • Replaced direct *github.Client usage in the service with the retrying wrapper.
  • Routed existing issue-listing, issue-create/edit, comment-create, and commit-fetch operations through the wrapper without changing higher-level flow.

• Targeted coverage

  • Added focused tests for retry-on-transient-error, stop-on-non-retryable-error, and context-cancel behavior.
  • Added retryability classification coverage for the GitHub and network error cases handled by the wrapper.

err := g.retry(ctx, "issues.create", func() error {
	var err error
	created, resp, err = g.doCreateIssue(ctx, owner, repo, issue)
	return err
})