Add retry controls to workflow log fetching

[Copilot]

Transient GitHub API failures could interrupt scheduled workflow diagnostics and leave log collection incomplete. Log fetching now supports bounded retries, backoff, timeouts, and recovery metadata.

• Request resilience

  • Add --retries, --backoff-seconds, and --timeout.
  • Retry HTTP 429 and 5xx responses with exponential backoff.
  • Honor Retry-After when provided.
  • Report retry attempts and terminal failures to stderr.

• Manifest diagnostics

  • Record per-run fetch attempts and final errors in manifest.json.

python3 scripts/fetch-workflow-logs.py ci.yml \
  --retries 5 \
  --backoff-seconds 2 \
  --timeout 45

• Fixes [product-manager-impersonator] Retry/backoff controls for workflow log fetching #1869