fix(gateway): 修复 Qwen3.8 推理参数并增加上游自愈

统一 Qwen3.8 固定思考、推理档位、预算互斥与温度下限规则，并覆盖 Chat Completions 和原生 Responses 请求。

新增安全参数纠正、有限重试和并发安全的进程内 LRU 学习；补充共享行为向量、httptest 回归与真实模型验收用例。

验证：go test ./... -count=1 通过，gofmt 检查通过；真实 Qwen3.8 流式、非流式及缓存重学习验收通过。

apps/api/internal/runner/service.go

@@ -154,6 +154,7 @@ func NewWithStores(
 	}
 	httpClients := newHTTPClientCache()
 	scriptExecutor := &scriptengine.Executor{Logger: logger}
+	openAIParameterCorrections := clients.NewParameterCorrectionCache()
 	service := &Service{
 		cfg:               cfg,
 		store:             db,
@@ -162,7 +163,7 @@ func NewWithStores(
 		logger:            logger,
 		scriptExecutor:    scriptExecutor,
 		clients: map[string]clients.Client{
-			"openai":                clients.OpenAIClient{HTTPClient: httpClients.none},
+			"openai":                clients.OpenAIClient{HTTPClient: httpClients.none, Corrections: openAIParameterCorrections},
 			"aliyun-bailian":        clients.AliyunBailianClient{HTTPClient: httpClients.none},
 			"blackforest":           clients.BlackforestClient{HTTPClient: httpClients.none},
 			"gemini":                clients.GeminiClient{HTTPClient: httpClients.none},
@@ -2562,6 +2563,11 @@ func validateRequest(kind string, body map[string]any) error {
 		if body["input"] == nil && body["messages"] == nil {
 			return errors.New("input or messages is required")
 		}
+		if reasoning, ok := body["reasoning"].(map[string]any); ok {
+			if err := clients.ValidateOpenAIReasoningEffort(reasoning["effort"]); err != nil {
+				return err
+			}
+		}
 		if value, explicit := nonNullParameter(body, "max_output_tokens"); explicit {
 			if _, ok := positiveInteger(value); !ok {
 				return &clients.ClientError{Code: "invalid_parameter", Message: "max_output_tokens must be a positive integer", Param: "max_output_tokens", StatusCode: 400, Retryable: false}