fix(asr): 一句话识别 sourceType=1 时支持显式 voiceFormat（test-0415 反馈3-1 续修）

- 老逻辑：req.setVoiceFormat(getVoiceFormat(request.getUrl()))
  sourceType=1 时 url 为空 → 默认 wav → 把 mp3 数据当 wav 给腾讯云 → "Audio decoding failed"
- 修复：TencentAsrRequest 新增 voiceFormat 字段（前端已透传 'mp3'）
  ToolTencentAsrServiceImpl.sentenceRecognition 优先使用 request.getVoiceFormat()，
  缺省回落到 getVoiceFormat(url) 兼容老调用

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

msh_crmeb_22/crmeb-common/src/main/java/com/zbkj/common/request/tencent/TencentAsrRequest.java

@@ -50,4 +50,7 @@ public class TencentAsrRequest {
 
     @ApiModelProperty(value = "是否显示词级别时间戳")
     private Boolean wordInfo;
+
+    @ApiModelProperty(value = "音频格式（sourceType=1 时必传：mp3/wav/m4a/aac/pcm/silk）", example = "mp3")
+    private String voiceFormat;
 }

msh_crmeb_22/crmeb-service/src/main/java/com/zbkj/service/service/impl/tool/ToolTencentAsrServiceImpl.java

@@ -179,7 +179,13 @@ public class ToolTencentAsrServiceImpl implements ToolTencentAsrService {
                     request.getEngineModelType() : asrConfig.getDefaultEngineModel());
             req.setSourceType(request.getSourceType() != null ?
                     request.getSourceType().longValue() : 0L);
-            req.setVoiceFormat(getVoiceFormat(request.getUrl()));
+            // test-0415 反馈3-1：sourceType=1 时 url 为空，从 url 推断会默认 wav，导致 mp3 数据被当 wav 解 → "Audio decoding failed"
+            // 优先使用前端显式传入的 voiceFormat
+            String voiceFormat = request.getVoiceFormat();
+            if (voiceFormat == null || voiceFormat.isEmpty()) {
+                voiceFormat = getVoiceFormat(request.getUrl());
+            }
+            req.setVoiceFormat(voiceFormat);
 
             if (request.getSourceType() != null && request.getSourceType() == 1) {
                 req.setData(request.getData());