feat(monitoring): comprehensive alerting overhaul, 66 rules across 14 PrometheusRules

The Longhorn memory leak went undetected for 14 days because alerting
was broken (email receiver, missing label selector, no node alerts).
This overhaul brings alerting to production grade.

Fixes:
- Alloy Loki URL pointed to deleted loki-gateway, now loki:3100
- seaweedfs-bucket-init crash on unsupported `mc versioning` command
- All PrometheusRules now have `release: kube-prometheus-stack` label
- Removed broken email receiver, Matrix-only alerting

New alert coverage:
- Node: memory, CPU, swap, filesystem, inodes, network, clock skew, OOM
- Kubernetes: deployment down, CronJob failed, pod crash-looping, PVC full
- Backups: Postgres barman stale/failed, WAL archiving, SeaweedFS mirror
- Observability: Prometheus WAL/storage/rules, Loki/Tempo/AlertManager down
- Services: Stalwart, Bulwark, Tuwunel, Sol, Valkey, OpenSearch (smart)
- SLOs: auth stack 99.9% burn rate, Matrix 99.5%, latency p95 < 2s
- Recording rules for Linkerd RED metrics and node aggregates
- Watchdog heartbeat → Matrix every 12h (dead pipeline detection)
- Inhibition: critical suppresses warning for same alert+namespace
- OpenSearchClusterYellow only fires with >1 data node (single-node aware)

base/monitoring/prometheus-values.yaml

@@ -61,7 +61,7 @@ grafana:
     - name: Loki
       type: loki
       uid: loki
-      url: "http://loki-gateway.monitoring.svc.cluster.local:80"
+      url: "http://loki.monitoring.svc.cluster.local:3100"
       access: proxy
       isDefault: false
       jsonData:
@@ -130,10 +130,6 @@ alertmanager:
             requests:
               storage: 2Gi
   config:
-    global:
-      smtp_from: "alerts@DOMAIN_SUFFIX"
-      smtp_smarthost: "postfix.lasuite.svc.cluster.local:25"
-      smtp_require_tls: false
     route:
       group_by: [alertname, namespace]
       group_wait: 30s
@@ -143,30 +139,26 @@ alertmanager:
       routes:
         - matchers:
             - alertname = Watchdog
-          receiver: "null"
+          receiver: matrix
+          repeat_interval: 12h
         - matchers:
             - severity = critical
-          receiver: critical
+          receiver: matrix
         - matchers:
             - severity = warning
           receiver: matrix
     receivers:
-      - name: "null"
-      - name: email
-        email_configs:
-          - to: "ops@DOMAIN_SUFFIX"
-            send_resolved: true
       - name: matrix
         webhook_configs:
           - url: "http://matrix-alertmanager-receiver.monitoring.svc.cluster.local:3000/alerts/alerts"
             send_resolved: true
-      - name: critical
-        webhook_configs:
-          - url: "http://matrix-alertmanager-receiver.monitoring.svc.cluster.local:3000/alerts/alerts"
-            send_resolved: true
-        email_configs:
-          - to: "ops@DOMAIN_SUFFIX"
-            send_resolved: true
+    inhibitRules:
+      # Critical alerts suppress warnings for the same alertname+namespace
+      - source_matchers:
+          - severity = critical
+        target_matchers:
+          - severity = warning
+        equal: [alertname, namespace]
 
 # Disable monitors for components k3s doesn't expose
 kubeEtcd: