feat(observability): ✨ add cluster heartbeat dead-man switch alerts

ClusterMetricsSilent: fires if no kubelet metrics for >10m (catches vmagent outages).
ClusterAPIServerDown: fires if apiserver scrape fails for >5m.
Replaces silenced KubeControllerManagerDown/KubeSchedulerDown which never fire on managed K8s.

otc/observability.buildth.ing/stacks/observability/victoria-k8s-stack/manifests/alerts.yaml

@@ -93,3 +93,27 @@ spec:
             summary: "Node memory >90% on {{ $labels.cluster_environment }}"
             description: "Node memory utilization above 90% for 10 minutes."
             value: "{{ $value | humanizePercentage }}"
+
+    - name: cluster-health
+      rules:
+        - alert: ClusterMetricsSilent
+          expr: |
+            count(up{job="kubelet"}) by (cluster_environment) < 1
+            or
+            absent(up{job="kubelet", cluster_environment="dev"})
+          for: 10m
+          labels:
+            severity: critical
+          annotations:
+            summary: "Cluster {{ $labels.cluster_environment }} stopped sending metrics"
+            description: "No kubelet metrics received from cluster {{ $labels.cluster_environment }} for over 10 minutes. Either vmagent is dead or the cluster is unreachable."
+
+        - alert: ClusterAPIServerDown
+          expr: |
+            up{job="apiserver", cluster_environment=~".+"} == 0
+          for: 5m
+          labels:
+            severity: critical
+          annotations:
+            summary: "API server down on {{ $labels.cluster_environment }}"
+            description: "Kubernetes API server scrape is failing on cluster {{ $labels.cluster_environment }}."