fix(traefik): size probes for HDD stalls

Single replica is the whole ingress; liveness kills at 2s timeouts took every public service down in a loop. Same stall-sized budgets as postgres/metallb. Applied live via helm (pinned 41.5.0, values from git).

traefik/k8s/traefik-values.yaml

@@ -46,6 +46,22 @@ resources:
     cpu: "1000m"
     memory: "1Gi"
 
+# Single replica is the whole ingress: a SIGKILL here takes every public
+# service down. Budgets are sized for HDD stalls on a loaded node, not for a
+# healthy disk — same treatment as immich postgres and metallb speaker.
+readinessProbe:
+  failureThreshold: 6
+  initialDelaySeconds: 10
+  periodSeconds: 10
+  successThreshold: 1
+  timeoutSeconds: 5
+livenessProbe:
+  failureThreshold: 6
+  initialDelaySeconds: 30
+  periodSeconds: 30
+  successThreshold: 1
+  timeoutSeconds: 5
+
 providers:
   kubernetesIngress:
     enabled: true