feat: 新增全局错误透传规则功能

支持管理员配置上游错误如何返回给客户端：
- 新增 ErrorPassthroughRule 数据模型和 Ent Schema
- 实现规则的 CRUD API（/admin/error-passthrough-rules）
- 支持按错误码、关键词匹配，支持 any/all 匹配模式
- 支持按平台过滤（anthropic/openai/gemini/antigravity）
- 支持透传或自定义响应状态码和错误消息
- 实现两级缓存（Redis + 本地内存）和多实例同步
- 集成到 gateway_handler 的错误处理流程
- 新增前端管理界面组件
- 新增单元测试覆盖核心匹配逻辑

优化：
- 移除 refreshLocalCache 中的冗余排序（数据库已排序）
- 后端 Validate() 增加匹配条件非空校验

backend/internal/handler/openai_gateway_handler.go

@@ -22,11 +22,12 @@ import (
 
 // OpenAIGatewayHandler handles OpenAI API gateway requests
 type OpenAIGatewayHandler struct {
-	gatewayService      *service.OpenAIGatewayService
-	billingCacheService *service.BillingCacheService
-	apiKeyService       *service.APIKeyService
-	concurrencyHelper   *ConcurrencyHelper
-	maxAccountSwitches  int
+	gatewayService          *service.OpenAIGatewayService
+	billingCacheService     *service.BillingCacheService
+	apiKeyService           *service.APIKeyService
+	errorPassthroughService *service.ErrorPassthroughService
+	concurrencyHelper       *ConcurrencyHelper
+	maxAccountSwitches      int
 }
 
 // NewOpenAIGatewayHandler creates a new OpenAIGatewayHandler
@@ -35,6 +36,7 @@ func NewOpenAIGatewayHandler(
 	concurrencyService *service.ConcurrencyService,
 	billingCacheService *service.BillingCacheService,
 	apiKeyService *service.APIKeyService,
+	errorPassthroughService *service.ErrorPassthroughService,
 	cfg *config.Config,
 ) *OpenAIGatewayHandler {
 	pingInterval := time.Duration(0)
@@ -46,11 +48,12 @@ func NewOpenAIGatewayHandler(
 		}
 	}
 	return &OpenAIGatewayHandler{
-		gatewayService:      gatewayService,
-		billingCacheService: billingCacheService,
-		apiKeyService:       apiKeyService,
-		concurrencyHelper:   NewConcurrencyHelper(concurrencyService, SSEPingFormatComment, pingInterval),
-		maxAccountSwitches:  maxAccountSwitches,
+		gatewayService:          gatewayService,
+		billingCacheService:     billingCacheService,
+		apiKeyService:           apiKeyService,
+		errorPassthroughService: errorPassthroughService,
+		concurrencyHelper:       NewConcurrencyHelper(concurrencyService, SSEPingFormatComment, pingInterval),
+		maxAccountSwitches:      maxAccountSwitches,
 	}
 }
 
@@ -201,7 +204,7 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 	maxAccountSwitches := h.maxAccountSwitches
 	switchCount := 0
 	failedAccountIDs := make(map[int64]struct{})
-	lastFailoverStatus := 0
+	var lastFailoverErr *service.UpstreamFailoverError
 
 	for {
 		// Select account supporting the requested model
@@ -213,7 +216,11 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 				h.handleStreamingAwareError(c, http.StatusServiceUnavailable, "api_error", "No available accounts: "+err.Error(), streamStarted)
 				return
 			}
-			h.handleFailoverExhausted(c, lastFailoverStatus, streamStarted)
+			if lastFailoverErr != nil {
+				h.handleFailoverExhausted(c, lastFailoverErr, streamStarted)
+			} else {
+				h.handleFailoverExhaustedSimple(c, 502, streamStarted)
+			}
 			return
 		}
 		account := selection.Account
@@ -278,12 +285,11 @@ func (h *OpenAIGatewayHandler) Responses(c *gin.Context) {
 			var failoverErr *service.UpstreamFailoverError
 			if errors.As(err, &failoverErr) {
 				failedAccountIDs[account.ID] = struct{}{}
+				lastFailoverErr = failoverErr
 				if switchCount >= maxAccountSwitches {
-					lastFailoverStatus = failoverErr.StatusCode
-					h.handleFailoverExhausted(c, lastFailoverStatus, streamStarted)
+					h.handleFailoverExhausted(c, failoverErr, streamStarted)
 					return
 				}
-				lastFailoverStatus = failoverErr.StatusCode
 				switchCount++
 				log.Printf("Account %d: upstream error %d, switching account %d/%d", account.ID, failoverErr.StatusCode, switchCount, maxAccountSwitches)
 				continue
@@ -324,7 +330,37 @@ func (h *OpenAIGatewayHandler) handleConcurrencyError(c *gin.Context, err error,
 		fmt.Sprintf("Concurrency limit exceeded for %s, please retry later", slotType), streamStarted)
 }
 
-func (h *OpenAIGatewayHandler) handleFailoverExhausted(c *gin.Context, statusCode int, streamStarted bool) {
+func (h *OpenAIGatewayHandler) handleFailoverExhausted(c *gin.Context, failoverErr *service.UpstreamFailoverError, streamStarted bool) {
+	statusCode := failoverErr.StatusCode
+	responseBody := failoverErr.ResponseBody
+
+	// 先检查透传规则
+	if h.errorPassthroughService != nil && len(responseBody) > 0 {
+		if rule := h.errorPassthroughService.MatchRule("openai", statusCode, responseBody); rule != nil {
+			// 确定响应状态码
+			respCode := statusCode
+			if !rule.PassthroughCode && rule.ResponseCode != nil {
+				respCode = *rule.ResponseCode
+			}
+
+			// 确定响应消息
+			msg := service.ExtractUpstreamErrorMessage(responseBody)
+			if !rule.PassthroughBody && rule.CustomMessage != nil {
+				msg = *rule.CustomMessage
+			}
+
+			h.handleStreamingAwareError(c, respCode, "upstream_error", msg, streamStarted)
+			return
+		}
+	}
+
+	// 使用默认的错误映射
+	status, errType, errMsg := h.mapUpstreamError(statusCode)
+	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
+}
+
+// handleFailoverExhaustedSimple 简化版本，用于没有响应体的情况
+func (h *OpenAIGatewayHandler) handleFailoverExhaustedSimple(c *gin.Context, statusCode int, streamStarted bool) {
 	status, errType, errMsg := h.mapUpstreamError(statusCode)
 	h.handleStreamingAwareError(c, status, errType, errMsg, streamStarted)
 }