feat(admin): Agent 并发 - 后台可调同时对话上限并展示 Agent 服务状态

新增运行策略 agentMaxSessions（默认 30，范围 1-64），保存后同时下发到内嵌进程闸门与独立 yingce-agent 的 /v1/limit，降低上限不会中断已开始的对话。系统性能页展示 Agent 服务健康、进行中/排队数量，并可直接修改上限。

backend/internal/agent/yingceagent/client.go

@@ -14,6 +14,68 @@ import (
 	agentruntime "infinite-canvas/backend/internal/agent/runtime"
 )
 
+type Status struct {
+	OK     bool `json:"ok"`
+	Active int  `json:"active"`
+	Limit  int  `json:"limit"`
+	Queued int  `json:"queued"`
+}
+
+func Inspect(ctx context.Context, endpoint string) (Status, error) {
+	endpoint = strings.TrimRight(strings.TrimSpace(endpoint), "/")
+	if endpoint == "" {
+		return Status{}, fmt.Errorf("YINGCE_AGENT_URL is empty")
+	}
+	request, err := http.NewRequestWithContext(ctx, http.MethodGet, endpoint+"/healthz", nil)
+	if err != nil {
+		return Status{}, err
+	}
+	response, err := remoteClient.Do(request)
+	if err != nil {
+		return Status{}, err
+	}
+	defer response.Body.Close()
+	if response.StatusCode != http.StatusOK {
+		return Status{}, fmt.Errorf("yingce-agent health returned HTTP %d", response.StatusCode)
+	}
+	var status Status
+	if err := json.NewDecoder(io.LimitReader(response.Body, 4096)).Decode(&status); err != nil {
+		return Status{}, err
+	}
+	return status, nil
+}
+
+func SetLimit(ctx context.Context, endpoint, serviceToken string, limit int) error {
+	endpoint = strings.TrimRight(strings.TrimSpace(endpoint), "/")
+	serviceToken = strings.TrimSpace(serviceToken)
+	if endpoint == "" {
+		return fmt.Errorf("YINGCE_AGENT_URL is empty")
+	}
+	if len(serviceToken) < 32 {
+		return fmt.Errorf("YINGCE_AGENT_TOKEN must contain at least 32 characters")
+	}
+	payload, err := json.Marshal(map[string]int{"maxSessions": limit})
+	if err != nil {
+		return err
+	}
+	request, err := http.NewRequestWithContext(ctx, http.MethodPut, endpoint+"/v1/limit", bytes.NewReader(payload))
+	if err != nil {
+		return err
+	}
+	request.Header.Set("Authorization", "Bearer "+serviceToken)
+	request.Header.Set("Content-Type", "application/json")
+	response, err := remoteClient.Do(request)
+	if err != nil {
+		return err
+	}
+	defer response.Body.Close()
+	if response.StatusCode != http.StatusOK {
+		detail, _ := io.ReadAll(io.LimitReader(response.Body, 1024))
+		return fmt.Errorf("yingce-agent limit returned HTTP %d: %s", response.StatusCode, strings.TrimSpace(string(detail)))
+	}
+	return nil
+}
+
 // Run executes one Agent session on the independent yingce-agent service.
 // The Go process keeps the bridge, billing, and tool execution. A configured
 // remote endpoint never falls back to an embedded Node process: doing so can

backend/internal/agent/yingceagent/client_test.go

@@ -116,4 +116,14 @@ if (!response.ok) {
 	if !called {
 		t.Fatal("remote runtime did not call the Go tool bridge")
 	}
+	if err := SetLimit(ctx, endpoint, token, 16); err != nil {
+		t.Fatal(err)
+	}
+	status, err := Inspect(ctx, endpoint)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if status.Limit != 16 {
+		t.Fatalf("limit = %d, want 16", status.Limit)
+	}
 }