[pr-staging-deploy]: Improve PR staging deploy comments: reliability, partial status, and Sliplane listing merge

[ArtemLazarchuk]

Summary

Improves the PR Staging Deploy sample so GitHub PR comments stay accurate and easier to reason about when Sliplane is slow or only one service succeeds.

Changes

• Immediate progress comments after a deploy/redeploy is queued (Deploy in progress, Redeploying staging, Checking staging deployment) so threads do not look stuck after e.g. “Deleted”.
• Comment refresh model: remove all prior marker comments (<!-- ivy-staging-deploy -->) with retries, then post a new comment each time (matches desired “always fresh” thread behavior).
• Status resolution: merge per-service state from Sliplane events + listing on every poll (not only after a long delay), so “pending” clears sooner when listing is already live.
• Partial deployments: if Docs and Samples diverge (one live, one failed, or one not provisioned), post a Partial deployment comment with per-service lines and logs focused on the failed side.
• URL gating: when posting “Deployed”, only require URLs for services that were actually created.
• GitHub API: treat 404 on comment delete as success (idempotent cleanup).

Notes

• End-to-end behavior still depends on Sliplane and GitHub availability; this reduces “silent” periods and misleading all-failed messaging when only one wing fails.