[pr-staging-deploy]: improve reliability of staging deploy lifecycle

[ArtemLazarchuk]

Summary

• No intermediate comments: bot no longer posts "Deploying..." or "Redeploying..." —
  only one final comment is created: either "Deployed" (with links) or "Deploy failed" (with logs)

• No comment editing: all final comments use forceNewComment: true to avoid
  unstable comment IDs and edit race conditions

• Fork PR support: deploy now uses the fork's clone_url from the webhook payload,
  so Sliplane clones the correct repository when a PR comes from a fork

• synchronize fallback: if new commits are pushed to a PR but no services exist yet
  (first push, services were deleted, or rapid consecutive pushes), falls back to a fresh
  deploy instead of reporting "Redeploy failed: 0 service(s)"

• Startup scan: on app restart, pending Sliplane deployments are detected and
  re-enqueued automatically — PR comments are updated to their final state after redeploy

• Hard timeout: polling loop stops after 30 minutes with a clear "Deploy timed out" comment
  instead of running indefinitely

• PR-number-based service naming: services are now named ivy-staging-docs-{prNumber}
  instead of using the sanitized branch name, making lookup deterministic across all events