Poll for exited container in prune integration test

Summary

TestDockerCommandWithDoingPruneAndAfterScript checked, in a single one-shot grep, whether an earlier build stage's runner-doprune-* container had already transitioned to the exited state. Observed failing with exit code 1 in CI on an unrelated MR: https://gitlab.com/gitlab-org/gitlab-runner/-/jobs/15212124101

Note: the exact trigger isn't confirmed — I don't have evidence this reproduces reliably, only a single observed failure. The fix below is a best-effort mitigation for the most likely cause (a race between the container's state transition and this check), not a confirmed root-cause fix.

Suspected cause

The container's transition to exited isn't necessarily visible via the shared docker socket the instant the earlier build stage's container stops, so a single immediate check can race with that transition.

Fix

Poll for up to 10 seconds instead of checking once, while still failing the job if the container is genuinely never found — preserving the test's original intent (it should still fail if prune functionality doesn't work).

Verification

• go build / go vet clean, golangci-lint 0 issues.
• Verified the retry-loop shell logic directly under alpine:latest's sh for both the "found on a later attempt" and "never found" cases — the latter still exits 1 as before.
• Could not reproduce the original race directly (it depends on runner-host timing), so this MR should be watched for whether the failure recurs after merge.

Closes #39591 (closed)