fix(cluster): 延长 Worker 启动探针保护期

Worker 在切换后恢复积压 River 任务时可能超过原 liveness 初始窗口，导致健康端口尚未监听就被误杀。增加最长 10 分钟 startupProbe，启动完成后仍沿用原 readiness 和 liveness。\n\n验证：kubectl kustomize、服务端 dry-run、生产滚动更新。

deploy/kubernetes/production/application.yaml

@@ -307,6 +307,13 @@ spec:
           ports:
             - name: health
               containerPort: 8088
+          startupProbe:
+            httpGet:
+              path: /healthz
+              port: health
+            periodSeconds: 5
+            timeoutSeconds: 3
+            failureThreshold: 120
           readinessProbe:
             httpGet:
               path: /readyz
@@ -425,6 +432,13 @@ spec:
           ports:
             - name: health
               containerPort: 8088
+          startupProbe:
+            httpGet:
+              path: /healthz
+              port: health
+            periodSeconds: 5
+            timeoutSeconds: 3
+            failureThreshold: 120
           readinessProbe:
             httpGet:
               path: /readyz