修复 Grok 孤儿控件、thinking 断言与流式 failover 的 CI 回归

畸形 tools 继续剥离孤儿 tool_choice；bare error 在临时上游失败时保持可换号；已转发的官方失败帧不再重复写入。

backend/internal/handler/openai_gateway_credential_failover_test.go

@@ -133,7 +133,8 @@ func TestResponsesFailoverExhaustedAfterForwardedTerminalMarksOpsWithoutDuplicat
 	recorder := httptest.NewRecorder()
 	c, _ := gin.CreateTestContext(recorder)
 	official := "event: response.failed\ndata: {\"type\":\"response.failed\",\"response\":{\"status\":\"failed\",\"error\":{\"code\":\"server_error\",\"message\":\"official failure\"}}}\n\n"
-	_, _ = recorder.Write([]byte(official))
+	_, err := c.Writer.Write([]byte(official))
+	require.NoError(t, err)
 	service.MarkOpsStreamError(c, "server_error", "official failure", http.StatusBadGateway)
 
 	(&GatewayHandler{}).handleResponsesFailoverExhausted(c, &service.UpstreamFailoverError{

backend/internal/service/openai_gateway_grok.go

@@ -984,9 +984,7 @@ func sanitizeGrokResponsesTools(body []byte) ([]byte, error) {
 		return deleteGrokOrphanToolControls(body)
 	}
 	if !tools.IsArray() {
-		// Leave malformed tools (and their tool_choice) untouched instead of
-		// inventing an empty inventory.
-		return body, nil
+		return deleteGrokOrphanToolControls(body)
 	}
 
 	rawTools := tools.Array()

backend/internal/service/openai_gateway_grok_chat_bridge_test.go

@@ -481,7 +481,7 @@ func TestForwardGrokChatViaResponsesTraeCompatibilityFieldsKeepCacheRoute(t *tes
 	require.Equal(t, "priority", gjson.GetBytes(upstream.lastBody, "service_tier").String())
 	require.False(t, gjson.GetBytes(upstream.lastBody, "stop").Exists())
 	require.False(t, gjson.GetBytes(upstream.lastBody, "reasoning").Exists())
-	require.Contains(t, gjson.GetBytes(upstream.lastBody, "input.1.content.0.text").String(), "<thinking>I should use lookup</thinking>")
+	require.Contains(t, gjson.GetBytes(upstream.lastBody, "input.1.content").String(), "<thinking>I should use lookup</thinking>")
 	require.Equal(t, "function_call", gjson.GetBytes(upstream.lastBody, "input.2.type").String())
 	require.Equal(t, "function_call_output", gjson.GetBytes(upstream.lastBody, "input.3.type").String())
 

backend/internal/service/openai_gateway_grok_test.go

@@ -418,10 +418,9 @@ func TestSanitizeGrokResponsesToolsKeepsToolChoiceOnlyWithSupportedTools(t *test
 			wantToolChoice: true,
 		},
 		{
-			name:           "malformed non-array tools remain untouched",
-			body:           `{"input":"hello","tools":{"type":"function","name":"lookup"},"tool_choice":"auto"}`,
-			wantTools:      true,
-			wantToolChoice: true,
+			name:      "malformed non-array tools drop orphan controls",
+			body:      `{"input":"hello","tools":{"type":"function","name":"lookup"},"tool_choice":"auto"}`,
+			wantTools: true,
 		},
 	}
 

backend/internal/service/openai_gateway_passthrough.go

@@ -1491,7 +1491,15 @@ func openAIStreamErrorEventShouldFailover(payload []byte, message string) bool {
 	case http.StatusUnauthorized, http.StatusTooManyRequests, 529:
 		return true
 	}
-	return isOpenAITransientProcessingError(http.StatusBadRequest, message, payload)
+	if isOpenAITransientProcessingError(http.StatusBadRequest, message, payload) {
+		return true
+	}
+	combined := strings.ToLower(strings.TrimSpace(message + " " +
+		gjson.GetBytes(payload, "error.message").String() + " " +
+		gjson.GetBytes(payload, "response.error.message").String()))
+	return strings.Contains(combined, "temporary") ||
+		strings.Contains(combined, "try again") ||
+		strings.Contains(combined, "please retry")
 }
 
 func (s *OpenAIGatewayService) handleOpenAIStreamTerminalAccountSideEffects(