Wait for Cloud Run routing convergence during promotion

[anth-volk]

Fixes #699

Summary

Cloud Run can continue routing requests to the previous revision briefly after its service status reports that 100% of traffic has moved. The production deployment workflow previously made one immediate health request and restored the previous revision when that request reached the old revision.

This change lets stable and custom hostname checks wait for that documented routing transition without hiding unrelated deployment failures.

Changes

• Pass the known previous revision to stable and custom hostname checks after promotion.
• Retry only when a healthy response identifies that exact previous revision.
• Use at most 24 attempts spaced five seconds apart by default.
• Fail immediately for unhealthy responses, missing revision headers, and unrelated revisions.
• Keep tagged candidate checks immediate and exact.
• Add automated coverage for convergence, timeout, unexpected revisions, candidate behavior, and workflow argument wiring.

Testing

• uv run pytest tests/test_deployment_assets.py -v — 25 passed
• shellcheck .github/scripts/cloud-run-simulation-entry-smoke.sh
• actionlint .github/workflows/simulation-deploy.reusable.yml
• git diff --check