fix: eliminate memory overcommit to prevent node OOM crashes

Set requests = limits (Guaranteed QoS) across LimitRange defaults and
explicit pod resources. Node2 crashed 2026-03-14 from 250% memory
overcommit (61GB limits on 24GB node).

Changes:
- LimitRange: default = defaultRequest for all 6 tiers
- Grafana: 3 → 2 replicas
- Grampsweb: document why replicas=0
- Prometheus: 1Gi/4Gi → 3Gi/3Gi
- OpenClaw: 512Mi/2Gi → 768Mi/768Mi
- Immich server: 256Mi/2Gi → 512Mi/512Mi
- Immich postgresql: 256Mi/1Gi → 512Mi/512Mi
- Calibre: 256Mi/1536Mi → 256Mi/256Mi
- Linkwarden: 256Mi/1536Mi → 768Mi/768Mi
- N8N: 256Mi/1Gi → 512Mi/512Mi
- MySQL cluster: 1Gi/3-4Gi → 2Gi/2Gi
- pg-cluster (CNPG): 512Mi/4Gi → 512Mi/512Mi
- DBaaS ResourceQuota limits.memory: 64Gi → 12Gi

[ci skip]

stacks/platform/modules/monitoring/grafana_chart_values.yaml

@@ -1,6 +1,6 @@
 deploymentStrategy:
   type: RollingUpdate
-replicas: 3
+replicas: 2
 adminPassword: "${grafana_admin_password}"
 resources:
   requests:

stacks/platform/modules/monitoring/prometheus_chart_values.tpl

@@ -141,9 +141,9 @@ server:
   resources:
     requests:
       cpu: 100m
-      memory: 1Gi
+      memory: 3Gi
     limits:
-      memory: 4Gi
+      memory: 3Gi
   strategy:
     type: Recreate
   baseURL: "https://prometheus.viktorbarzin.me"