fix: prevent build watch from redeploying old builds on restart

[aperolschpritz]

Problem

When QuickStack restarts, the build watch redeploys old successful builds.

1. Builds that were already built and deployed a long time ago get deployed again.
2. Wrong/older versions get deployed after restart, because with multiple successful jobs per app an older one can win.

Root cause

k8s.Watch.watch(path, {}, ...) is started without a resourceVersion. Per the Kubernetes API, an unset resourceVersion uses "Get State and Start at Most Recent": the watch begins with synthetic ADDED events for every existing build Job.

handleJobEvent only checks the in-memory processedJobs set, which is empty after a process restart. Every replayed SUCCEEDED job therefore triggers handleSucceeded -> createDeployment. Jobs are replayed in list order (effectively alphabetical by job name), so among several successful builds the older one can be processed last and overwrite the deployment. This is issue #100.

A previous scanExistingJobs guard was removed in 8bd76dc, leaving the watch replay unguarded.

Fix

V1: seed processedJobs with existing terminal jobs before the watch starts.

• seedProcessedJobs() lists jobs and marks SUCCEEDED/FAILED as processed. Running/pending jobs stay unseeded so they still deploy when they finish.
• Replayed ADDED events for old builds are now ignored.
• handleJobEvent also ignores DELETED events, so cleaning up finished jobs cannot trigger a deployment.

Tests

• Old successful build replayed as ADDED after startup is not redeployed.
• A running build is still deployed once it succeeds after startup.
• DELETED events are ignored even when the job shows as succeeded.

Verified locally: unit tests pass, ESLint clean, tsc --noEmit --project tsconfig.server.json clean.