feat: 十轮网关优化 - 安全加固/可观测性/性能/可靠性

- SSE Keepalive Ping (15s心跳防止代理断连)
- Timing HTTP 头 (X-Timing-Queue/Inference/Total-Ms)
- Adapter Request-ID 传播到后端
- Session 清理日志回调
- Server 安全加固 (ReadHeaderTimeout/MaxHeaderBytes 防 slowloris)
- Usage Tracker 数据保留清理 (retentionDays + 定期清理)
- Config Reload 后 Adapter Registry 更新 (RegisterIfAbsent + RWMutex)
- Rate Limiter 空闲 Bucket 清理 (30分钟过期)
- Shutdown Drain 超时可配置 (ShutdownDrainSeconds)
- Config 模型字段校验增强 (provider/endpoint/actual_model)
- Auth 过期 Key 自动清理 (5分钟扫描)
- Admin API Rate Limiting
- Adapter Health Check 独立超时 (每个 adapter 3s)
- TCP 连接阶段超时 (DialContext 5s + KeepAlive 30s)
- 幂等键缓存、审计日志、Gzip 中间件、CORS Expose Headers
- Backpressure 响应头、熔断器 Prometheus 指标
- 连接池优化、Trace-ID 全链路传播

internal/adapter/vllm.go

@@ -24,10 +24,8 @@ type VLLMAdapter struct {
 // NewVLLMAdapter creates a new vLLM adapter.
 func NewVLLMAdapter(endpoint string) *VLLMAdapter {
 	return &VLLMAdapter{
-		endpoint: strings.TrimRight(endpoint, "/"),
-		httpClient: &http.Client{
-			Timeout: 120 * time.Second,
-		},
+		endpoint:   strings.TrimRight(endpoint, "/"),
+		httpClient: newHTTPClient(120 * time.Second),
 	}
 }
 
@@ -37,12 +35,12 @@ func (a *VLLMAdapter) Name() string {
 
 // vllmChatRequest is the OpenAI-compatible chat request for vLLM.
 type vllmChatRequest struct {
-	Model       string      `json:"model"`
-	Messages    []vllmMsg   `json:"messages"`
-	Stream      bool        `json:"stream"`
-	MaxTokens   int         `json:"max_tokens,omitempty"`
-	Temperature *float64    `json:"temperature,omitempty"`
-	TopP        *float64    `json:"top_p,omitempty"`
+	Model       string    `json:"model"`
+	Messages    []vllmMsg `json:"messages"`
+	Stream      bool      `json:"stream"`
+	MaxTokens   int       `json:"max_tokens,omitempty"`
+	Temperature *float64  `json:"temperature,omitempty"`
+	TopP        *float64  `json:"top_p,omitempty"`
 }
 
 type vllmMsg struct {
@@ -55,9 +53,9 @@ type vllmChatResponse struct {
 	ID      string `json:"id"`
 	Model   string `json:"model"`
 	Choices []struct {
-		Index        int    `json:"index"`
+		Index        int     `json:"index"`
 		Message      vllmMsg `json:"message"`
-		FinishReason string `json:"finish_reason"`
+		FinishReason string  `json:"finish_reason"`
 	} `json:"choices"`
 	Usage struct {
 		PromptTokens     int `json:"prompt_tokens"`
@@ -71,7 +69,7 @@ type vllmStreamChunk struct {
 	ID      string `json:"id"`
 	Model   string `json:"model"`
 	Choices []struct {
-		Index        int    `json:"index"`
+		Index        int     `json:"index"`
 		Delta        vllmMsg `json:"delta"`
 		FinishReason *string `json:"finish_reason"`
 	} `json:"choices"`
@@ -95,6 +93,9 @@ func (a *VLLMAdapter) ChatCompletion(ctx context.Context, req *ChatRequest) (*Ch
 		return nil, fmt.Errorf("create vllm request: %w", err)
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
+	if req.RequestID != "" {
+		httpReq.Header.Set("X-Request-ID", req.RequestID)
+	}
 
 	resp, err := a.httpClient.Do(httpReq)
 	if err != nil {
@@ -145,6 +146,9 @@ func (a *VLLMAdapter) ChatCompletionStream(ctx context.Context, req *ChatRequest
 	}
 	httpReq.Header.Set("Content-Type", "application/json")
 	httpReq.Header.Set("Accept", "text/event-stream")
+	if req.RequestID != "" {
+		httpReq.Header.Set("X-Request-ID", req.RequestID)
+	}
 
 	resp, err := a.httpClient.Do(httpReq)
 	if err != nil {