fix(pool): ignore scale-set job events

[rm-you]

Summary

Scale-set jobs reach the pool manager through database watcher events, even without webhooks. The startup query already filters on workflow_job_id > 0, but the watcher caches every non-completed job belonging to the entity. As a result, queued scale-set jobs are incorrectly considered for pool allocation.

For scale-set workflows using runs-on: group: without labels, each pool reconciliation logs error finding pools matching labels: missing tags. Scale-set jobs with labels also enter the pool-matching path, despite belonging to the scale-set worker.

Apply the same positive workflow-job-ID check to incoming watcher events. This leaves normal webhook handling unchanged and does not mask empty-label errors for webhook jobs. No database changes or dependencies on other PRs.

Tests

• Table-driven coverage of create/update events for labeled and unlabeled scale-set and webhook jobs, plus missing/invalid workflow IDs.
• SQLite-backed regression showing that the startup query and live watcher both queue only the webhook job when both job types are stored.
• Confirmed the new regressions fail on unmodified upstream and pass with the fix.
• GOWORK=off go test -race -mod=vendor -tags testing -timeout=15m -parallel=4 -count=1 ./...
• Pinned golangci-lint v2.13.2 with --build-tags=testing,integration; formatting and git diff --check.