fix(worker): 将 Gemini 同步请求交由异步队列执行

将非流式 Gemini generateContent 改为 River Worker 执行，API 通过批量状态查询等待完成，避免高并发同步请求耗尽 API 数据库连接池。\n\n生成图片在持久化时保存带哈希的资产引用，响应恢复时下载并校验后重建 Base64，数据库不保存媒体原文。请求体物化增加前置内存门禁，并扩展双 API/Worker PostgreSQL 压力测试。\n\n验证：go test ./... -count=1；go vet ./...；pnpm openapi；迁移安全检查；64 与 256 请求双角色异步 Gemini 压力测试。

apps/api/internal/httpapi/gemini_compat.go

@@ -49,7 +49,7 @@ var geminiGenerateContentRoutePrefixes = []string{
 func (s *Server) registerGeminiGenerateContentRoutes(mux *http.ServeMux) {
 	handler := s.requireProtocolUser(clients.ProtocolGeminiGenerateContent, http.HandlerFunc(s.geminiGenerateContent))
 	for _, prefix := range geminiGenerateContentRoutePrefixes {
-		mux.Handle("POST "+prefix, geminiGenerateContentRouteHandler(prefix, handler))
+		mux.Handle("POST "+prefix, geminiGenerateContentRouteHandler(prefix, s.withMediaRequestBodySlot(handler)))
 	}
 }
 
@@ -122,7 +122,7 @@ func (s *Server) geminiGenerateContent(w http.ResponseWriter, r *http.Request) {
 		writeTaskTrafficError(w, err, clients.ProtocolGeminiGenerateContent)
 		return
 	}
-	releaseRequestBody, err := s.acquireMediaRequestBodySlot(r.Context())
+	releaseRequestBody, err := s.requestMediaBodyRelease(r.Context())
 	if err != nil {
 		return
 	}
@@ -181,7 +181,7 @@ func (s *Server) geminiGenerateContent(w http.ResponseWriter, r *http.Request) {
 		Model:           mapping.Model,
 		RunMode:         runMode,
 		AcceptanceRunID: admission.AcceptanceRunID,
-		Async:           false,
+		Async:           !streamMode,
 		Request:         prepared.Body,
 		ConversationID:  prepared.ConversationID,
 		NewMessageCount: prepared.NewMessageCount,
@@ -240,27 +240,36 @@ func (s *Server) geminiGenerateContent(w http.ResponseWriter, r *http.Request) {
 		s.writeGeminiGenerateContentStream(runCtx, w, r, task, user, mapping.Model, writeGeminiTaskError)
 		return
 	}
-	result, runErr := s.runner.Execute(runCtx, task, user)
-	if runErr != nil {
+	if err := s.runner.SubmitAsyncTask(runCtx, task); err != nil {
 		if !requestStillConnected(r) {
 			return
 		}
-		applyRunErrorHeaders(w, runErr)
-		if wire := clients.ErrorWireResponse(runErr); wireResponseMatches(wire, clients.ProtocolGeminiGenerateContent) {
-			writeWireResponse(w, wire)
-			return
-		}
-		writeGeminiTaskError(statusFromRunError(runErr), runErrorMessage(runErr), runErrorDetails(runErr), runErrorCode(runErr))
+		applyRunErrorHeaders(w, err)
+		writeGeminiTaskError(statusFromRunError(err), runErrorMessage(err), runErrorDetails(err), runErrorCode(err))
+		return
+	}
+	task, err = s.runner.WaitForTaskCompletion(runCtx, task.ID)
+	if err != nil || !requestStillConnected(r) {
+		return
+	}
+	if task.Status != "succeeded" {
+		status := storedTaskErrorStatus(task.ErrorCode)
+		writeGeminiTaskError(status, firstNonEmpty(task.ErrorMessage, task.Error, task.Message, "task failed"), nil, firstNonEmpty(task.ErrorCode, task.Status))
+		return
+	}
+	task, err = s.hydrateTaskResult(runCtx, task)
+	if err != nil {
+		writeGeminiTaskError(statusFromRunError(err), err.Error(), nil, clients.ErrorCode(err))
 		return
 	}
 	if !requestStillConnected(r) {
 		return
 	}
-	if wireResponseMatches(result.Wire, clients.ProtocolGeminiGenerateContent) {
-		writeWireResponse(w, result.Wire)
+	if _, native := task.Result["candidates"]; native {
+		writeJSON(w, http.StatusOK, task.Result)
 		return
 	}
-	writeJSON(w, http.StatusOK, geminiGenerateContentResponse(result.Output, mapping.Model))
+	writeJSON(w, http.StatusOK, geminiGenerateContentResponse(task.Result, mapping.Model))
 }
 
 func (s *Server) writeGeminiGenerateContentStream(runCtx context.Context, w http.ResponseWriter, r *http.Request, task store.GatewayTask, user *auth.User, model string, writeGeminiTaskError func(int, string, map[string]any, string)) {