ci: retry NVHPC image pull on transient nvcr.io timeouts

[sbryngelson]

Problem

NVHPC matrix jobs sporadically fail at the Pull NVHPC container step with:

Error response from daemon: Get "https://nvcr.io/v2/": context deadline exceeded

Example: NVHPC 25.7 (cpu) in run 27312594672, where 1 of 30 NVHPC jobs hit the timeout while the other 29 pulled the same registry fine — a transient nvcr.io flake, likely aggravated by ~30 concurrent pulls per run.

Fix

Retry docker pull up to 5 times with linear backoff (30s, 60s, ...). Docker resumes already-downloaded layers, so retries are cheap. Worst case adds ~5 min before failing with an explicit error annotation.

[Copilot]

Copilot encountered an error and was unable to review this pull request. You can try again by re-requesting a review.

[github-actions]

Claude Code Review

Head SHA: c8995c4

Files changed:

• 1
• .github/workflows/test.yml

Findings:

• Unnecessary 150-second sleep on final failed attempt (.github/workflows/test.yml): On the last loop iteration (attempt=5), when docker pull fails, the script falls through to sleep $((attempt * 30)) — sleeping 150 seconds — before the loop exits and the exit 1 fires. There is no further retry after iteration 5, so this sleep serves no purpose and adds 2.5 minutes to every eventual failure path. The echo message on that iteration also says "retrying in 150s…" which is misleading since no retry follows. Fix: skip the sleep (and misleading message) on the final attempt:

  for attempt in 1 2 3 4 5; do
      docker pull "$NVHPC_IMAGE" && exit 0
      if [ "$attempt" -lt 5 ]; then
          echo "docker pull failed (attempt $attempt/5); retrying in $((attempt * 30))s..."
          sleep $((attempt * 30))
      fi
  done
  echo "::error::Failed to pull $NVHPC_IMAGE after 5 attempts"
  exit 1