perf(queue): 限制大媒体任务内存并消除准入惊群

将同步与异步非文本任务的准入唤醒改为按任务和全局队首推进，批量续租等待者，避免大量请求同时争抢 advisory lock 和数据库连接。

对 Base64 请求解析、素材解码、上游媒体执行和结果物化增加分层并发限制，复用请求素材并释放重复 Gemini wire 数据；生产 API 默认入口解析并发 16，媒体物化并发 8。

新增 1000 个同步 Gemini 图像编辑请求的模拟上游压力验收，10 MiB 输入和输出下全部成功，Heap 峰值增长约 2.58 GiB，并验证共享上传哈希、单次 attempt 与本地零落盘。

验证：go test ./... -count=1；go vet ./...；gofmt -l 无输出；kubectl kustomize deploy/kubernetes/production；10 MiB Gemini Base64 千任务压力测试通过。

apps/api/internal/clients/helpers.go

@@ -65,12 +65,14 @@ func decodeHTTPResponseForProtocol(resp *http.Response, protocol string) (map[st
 		Protocol:   strings.TrimSpace(protocol),
 		StatusCode: resp.StatusCode,
 		Headers:    compatibleResponseHeaders(resp.Header),
-		RawJSON:    append([]byte(nil), raw...),
-	}
-	if len(raw) > 0 {
-		_ = json.Unmarshal(raw, &wire.Body)
+		// raw is already owned by this response. Reusing it avoids retaining a
+		// second copy of media responses that may be as large as 128 MiB.
+		RawJSON: raw,
 	}
 	if readErr != nil {
+		if len(raw) > 0 {
+			_ = json.Unmarshal(raw, &wire.Body)
+		}
 		return nil, wire, &ClientError{
 			Code:       "response_read_error",
 			Message:    readErr.Error(),
@@ -81,6 +83,9 @@ func decodeHTTPResponseForProtocol(resp *http.Response, protocol string) (map[st
 		}
 	}
 	if tooLarge {
+		if len(raw) > 0 {
+			_ = json.Unmarshal(raw, &wire.Body)
+		}
 		return nil, wire, &ClientError{
 			Code:       "response_too_large",
 			Message:    fmt.Sprintf("upstream JSON response exceeds %d bytes", maxBytes),
@@ -91,6 +96,9 @@ func decodeHTTPResponseForProtocol(resp *http.Response, protocol string) (map[st
 		}
 	}
 	if resp.StatusCode < 200 || resp.StatusCode >= 300 {
+		if len(raw) > 0 {
+			_ = json.Unmarshal(raw, &wire.Body)
+		}
 		return nil, wire, &ClientError{
 			Code:       statusCodeName(resp.StatusCode),
 			Message:    errorMessage(raw, resp.Status),
@@ -108,7 +116,10 @@ func decodeHTTPResponseForProtocol(resp *http.Response, protocol string) (map[st
 	if err := json.Unmarshal(raw, &out); err != nil {
 		return nil, wire, &ClientError{Code: "invalid_response", Message: err.Error(), StatusCode: resp.StatusCode, Retryable: false, Wire: wire}
 	}
-	wire.Body = cloneBody(out)
+	// Result and wire metadata are read-only until the runner replaces the
+	// canonical result, so they can share the decoded object instead of keeping
+	// another full copy of a large Base64 media response.
+	wire.Body = out
 	return out, wire, nil
 }