fix(capacity): 忽略未就绪的 Worker 节点

香港站点保留了已离线的旧 control-plane 节点标签，容量控制器仍尝试读取其 metrics，导致 404 并阻断 readiness。容量采集现在只纳入 Ready=True 的候选节点，实际 Worker Pod 所在的就绪节点继续参与容量计算。

验证：容量控制器聚焦测试与 API 全量 go test 均通过，并覆盖同站点 NotReady 旧节点与 Ready Worker 节点并存场景。

apps/api/internal/capacitycontroller/kubernetes.go

@@ -178,6 +178,20 @@ func (client *KubernetesClient) SiteState(ctx context.Context, site string) (Kub
 	if err := client.getJSON(ctx, nodePath, &nodes); err != nil {
 		return KubernetesSiteState{}, err
 	}
+	readyNodes := nodes.Items[:0]
+	for _, node := range nodes.Items {
+		ready := false
+		for _, condition := range node.Status.Conditions {
+			if condition.Type == "Ready" && condition.Status == "True" {
+				ready = true
+				break
+			}
+		}
+		if ready {
+			readyNodes = append(readyNodes, node)
+		}
+	}
+	nodes.Items = readyNodes
 	if len(nodes.Items) == 0 {
 		// A zero-replica Deployment with no eligible nodes represents an
 		// explicitly disabled site. It must remain visible to the planner, but it
@@ -186,7 +200,7 @@ func (client *KubernetesClient) SiteState(ctx context.Context, site string) (Kub
 		if state.CurrentReplicas == 0 {
 			return state, nil
 		}
-		return KubernetesSiteState{}, fmt.Errorf("site %s has no matching nodes", site)
+		return KubernetesSiteState{}, fmt.Errorf("site %s has no ready matching nodes", site)
 	}
 	var pods struct {
 		Items []struct {