feat(openai): 极致优化 OAuth 链路并补齐性能守护

- 优化 /v1/responses 热路径，减少重复解析与不必要拷贝\n- 优化并发与 token 竞争路径并补齐运行指标\n- 补充 OpenAI/Ops 相关单元测试与回归用例\n- 新增灰度阈值守护与压测脚本，支撑发布验收

backend/internal/handler/ops_error_logger.go

@@ -507,6 +507,7 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 				RetryCount:  0,
 				CreatedAt:   time.Now(),
 			}
+			applyOpsLatencyFieldsFromContext(c, entry)
 
 			if apiKey != nil {
 				entry.APIKeyID = &apiKey.ID
@@ -618,6 +619,7 @@ func OpsErrorLoggerMiddleware(ops *service.OpsService) gin.HandlerFunc {
 			RetryCount:  0,
 			CreatedAt:   time.Now(),
 		}
+		applyOpsLatencyFieldsFromContext(c, entry)
 
 		// Capture upstream error context set by gateway services (if present).
 		// This does NOT affect the client response; it enriches Ops troubleshooting data.
@@ -746,6 +748,44 @@ func extractOpsRetryRequestHeaders(c *gin.Context) *string {
 	return &s
 }
 
+func applyOpsLatencyFieldsFromContext(c *gin.Context, entry *service.OpsInsertErrorLogInput) {
+	if c == nil || entry == nil {
+		return
+	}
+	entry.AuthLatencyMs = getContextLatencyMs(c, service.OpsAuthLatencyMsKey)
+	entry.RoutingLatencyMs = getContextLatencyMs(c, service.OpsRoutingLatencyMsKey)
+	entry.UpstreamLatencyMs = getContextLatencyMs(c, service.OpsUpstreamLatencyMsKey)
+	entry.ResponseLatencyMs = getContextLatencyMs(c, service.OpsResponseLatencyMsKey)
+	entry.TimeToFirstTokenMs = getContextLatencyMs(c, service.OpsTimeToFirstTokenMsKey)
+}
+
+func getContextLatencyMs(c *gin.Context, key string) *int64 {
+	if c == nil || strings.TrimSpace(key) == "" {
+		return nil
+	}
+	v, ok := c.Get(key)
+	if !ok {
+		return nil
+	}
+	var ms int64
+	switch t := v.(type) {
+	case int:
+		ms = int64(t)
+	case int32:
+		ms = int64(t)
+	case int64:
+		ms = t
+	case float64:
+		ms = int64(t)
+	default:
+		return nil
+	}
+	if ms < 0 {
+		return nil
+	}
+	return &ms
+}
+
 type parsedOpsError struct {
 	ErrorType string
 	Message   string