fix: switch health probes from HTTP /health to TCP socket check

Most user apps (e.g. Next.js) don't have a /health endpoint, causing
liveness/readiness probes to fail with 404 and crash-loop the pods.
TCP socket probes just check if the port is open, which works universally.

backend/src/kubernetes/kubernetes.service.ts

@@ -216,14 +216,15 @@ export class KubernetesService implements OnModuleInit {
                   limits: { cpu: ctx.cpuLimit, memory: ctx.memoryLimit },
                 },
                 readinessProbe: {
-                  httpGet: { path: '/health', port: ctx.port as any },
-                  initialDelaySeconds: 10,
+                  tcpSocket: { port: ctx.port as any },
+                  initialDelaySeconds: 5,
                   periodSeconds: 5,
                 },
                 livenessProbe: {
-                  httpGet: { path: '/health', port: ctx.port as any },
-                  initialDelaySeconds: 30,
+                  tcpSocket: { port: ctx.port as any },
+                  initialDelaySeconds: 15,
                   periodSeconds: 10,
+                  failureThreshold: 5,
                 },
               },
             ],