feat: add cascade delete observability metrics

[ishtoo1]

What type of PR is this? (check all applicable)

• Feature

What changed?

• Add four Prometheus instruments to go/components/pipeline/metrics.go:
  • pipeline_cascade_delete_started_total{namespace, pipeline}
  • pipeline_cascade_delete_completed_total{namespace, pipeline}
  • pipeline_cascade_delete_error_total{namespace, pipeline, reason} (reasons: list_error, delete_error, update_error, kill_timeout)
  • pipeline_cascade_delete_active_children{namespace, pipeline, kind} (kinds: trigger_run, pipeline_run)
• Emit metrics at each phase transition and error path in handleDeletion.
• Add cascade-delete-started-at annotation so the "started" counter fires exactly once per cascade (not once per requeue).
• Add cascadeDeleteKillTimeout = 30m. After the timeout, controller stops killing and forcefully deletes child CRs with reason=\"kill_timeout\" on the error counter, so the Pipeline CR never stays stuck in Terminating forever.
• Add TestCascadeDelete_KillTimeout and TestCascadeDelete_CounterIncrementsOncePerCascade.

Why?
Addressing #1091.

• Observability for cascade-delete operations in dashboards / SLOs.
• Bound the controller's worst-case behavior when child workflows can't reach terminal state.

How did you test it?

• bazel test //go/components/pipeline/... — all tests pass.

• bazel build //go/... — no build errors.

• End-to-end behavior of the full cascade-delete stack is verified on a sandbox cluster; results are attached on docs: document pipeline cascade delete behavior #1114.

Potential risks

• New metric series. Cardinality is namespace × pipeline × (reason|kind). For typical deployments this is bounded; for operators managing thousands of pipelines the gauge may grow. Uses the same pattern as go/components/pipelinerun/metrics.go.
• Forceful delete after 30 minutes: a TR/PR CR may be removed while its underlying workflow is still running. This is intentional — the alternative is holding the pipeline CR in Terminating indefinitely. Operators can investigate stuck workflows via logs and the kill_timeout metric.

Release notes
New metrics emitted under the pipeline_cascade_delete_* prefix. No user-facing behavior change beyond the 30-minute kill timeout fallback.

Documentation Changes
N/A.

---

Stacked on top of #1112.

[CLAassistant]

All committers have signed the CLA.

[github-actions]

Go Coverage Report (Bazel)

Total Coverage: 63.4%

Coverage Policy:

• Baseline (existing code): ≥60% (current coverage)
• New/changed code: ≥90% ✅ STRICTLY ENFORCED
• Long-term goal: Improve baseline to 90%

View detailed HTML report in artifacts