Queue instead of cancel in the docs-deploy concurrency group so a stale
run that already passed check-freshness can never kill a fresher run's
in-progress deploy before skipping itself as stale. Also extract the
freshness-check shell logic into a standalone, tested script so the
dispatch-bypass and output contract can't regress silently.
Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
