fix: force Content-Type to application/json on non-streaming responses

When the gateway forces streaming to the upstream, the upstream returns
Content-Type: text/event-stream. In the non-streaming branches, the body
is correctly aggregated into JSON, but WriteFilteredHeaders has already
passed the upstream's text/event-stream header through. Since Gin's
c.JSON/c.Data only set Content-Type via writeContentType when it is not
already present, the SSE header is never overwritten -- so the body is JSON
while the header says SSE.

Downstream middlewares that decide "is this streaming?" by Content-Type
(e.g. new-api) then mis-handle the response as SSE, breaking non-streaming
JSON clients.

Fix: explicitly Set Content-Type to application/json right after
WriteFilteredHeaders in the three non-streaming JSON write paths.

backend/internal/service/gateway_forward_as_chat_completions.go

@@ -318,6 +318,12 @@ func (s *GatewayService) handleCCBufferedFromAnthropic(
 	if s.responseHeaderFilter != nil {
 		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
 	}
+	// 非流式响应必须是 application/json。上游被强制流式后会返回
+	// Content-Type: text/event-stream，经 WriteFilteredHeaders 透传后会污染
+	// 响应头；而 c.Data/c.JSON 走 Gin 的 writeContentType（仅当头不存在时才设置），
+	// 无法覆盖已存在的 SSE 头。这里显式 Set 强制改回 JSON，避免下游中间层
+	// （如 new-api）按 Content-Type 误判为流式。
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
 	// Marshal then bytes-replace so tool name mapping is reversed at byte level
 	// (parity with Parrot non-stream flow that marshals → restore → emit).
 	if respBytes, err := json.Marshal(ccResp); err == nil {

backend/internal/service/gateway_forward_as_responses.go

@@ -337,6 +337,12 @@ func (s *GatewayService) handleResponsesBufferedStreamingResponse(
 	if s.responseHeaderFilter != nil {
 		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
 	}
+	// 非流式响应必须是 application/json。上游被强制流式后会返回
+	// Content-Type: text/event-stream，经 WriteFilteredHeaders 透传后会污染
+	// 响应头；而 c.Data/c.JSON 走 Gin 的 writeContentType（仅当头不存在时才设置），
+	// 无法覆盖已存在的 SSE 头。这里显式 Set 强制改回 JSON，避免下游中间层
+	// （如 new-api）按 Content-Type 误判为流式。
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
 	if respBytes, err := json.Marshal(responsesResp); err == nil {
 		respBytes = reverseToolNamesIfPresent(c, respBytes)
 		c.Data(http.StatusOK, "application/json; charset=utf-8", respBytes)

backend/internal/service/openai_gateway_chat_completions.go

@@ -407,6 +407,11 @@ func (s *OpenAIGatewayService) handleChatBufferedStreamingResponse(
 	if s.responseHeaderFilter != nil {
 		responseheaders.WriteFilteredHeaders(c.Writer.Header(), resp.Header, s.responseHeaderFilter)
 	}
+	// 非流式响应必须为标准 JSON。上游被强制流式，其响应头 Content-Type 为
+	// text/event-stream，会经 WriteFilteredHeaders 透传进来；而 c.JSON 走 Gin 的
+	// writeContentType 仅在头不存在时才设置，无法覆盖。这里显式 Set 强制改回 JSON，
+	// 否则下游"看头判流式"的中间层（如 new-api）会把本应聚合的 JSON 当成 SSE 处理。
+	c.Writer.Header().Set("Content-Type", "application/json; charset=utf-8")
 	c.JSON(http.StatusOK, chatResp)
 
 	return &OpenAIForwardResult{