fix: surface real error cause and make round-robin summary visible in prod

[ttomalak-bkpr]

Summary

• GitHubRestStream.get_records now logs the real exception (type + message) before re-raising. singer_sdk's own catch-all only logs "an unhandled error occurred while syncing '<stream>'" with zero detail, so prod runs currently give no way to tell why a stream died (rate limit, auth, transient API error, ...). FatalAPIError/RetriableAPIError already carry a rich message (HTTP status, GitHub-Request-Id, etc.) - this just makes sure it isn't lost.
• The round-robin checkpoint's per-run summary (RepositoryStream._get_records_round_robin) was logged at info, but prod actually runs at warning - so despite already being deployed, this summary has never appeared in production logs. Bumped both the normal summary and the "0 repos returned" early-return to warning, so enumeration progress (repos processed, full lap or not, next resume point) is finally visible without any deploy-side log-level change.

This is scoped entirely to this repo - no changes to the bi-meltano launcher/image that runs it.

Test plan

• ruff check / ruff format --check on both changed files - clean (pre-existing unrelated line-length issues on repository_streams.py left untouched, confirmed present on master too)
• mypy on both changed files - clean
• Full existing test suite (pytest tests/) - same 72 passed / 10 pre-existing network-dependent failures as on master, no regressions