Add retry logic for Coveralls uploads to handle transient 503s

[intagliated]

Motivation and Context / Related issue

CI jobs (`unittest_py310_torch_release`, `unittest_py39_torch_release`, `integrationtest_py39_torch_release_cpu`) were intermittently failing at the "Send coverage to Coveralls (parallel)" step due to transient 503 Service Unavailable errors from Coveralls ("website is under heavy load, queue full"). These failures are unrelated to actual test or coverage results and were causing red CI runs on otherwise passing code.

This PR replaces the `coverallsapp/github-action@v2` step with the `coveralls` CLI wrapped in a retry loop (up to 4 attempts, 30s backoff), for both the parallel upload steps and the final `finish_coveralls_parallel` webhook call, so transient Coveralls outages no longer fail CI.

How Has This Been Tested

Workflow YAML validated for syntax. The retry logic itself will be exercised on this PR's own CI run — if Coveralls is healthy, the loop succeeds on the first attempt as before; if a 503 occurs, it will retry rather than fail immediately.

Checklist

• Documentation up-to-date — check it (no doc changes needed for a CI-only fix)
• CONTRIBUTING/CLA — check if already completed from your prior PRs
• All tests passed — leave unchecked for now, check it once you see this PR's CI run go green

[meta-codesync]

This pull request has been imported. If you are a Meta employee, you can view this in D114992440. (Because this pull request was imported automatically, there will not be any future comments.)