feat: 语音确认、联调与运维增强

- 语音：序数解析（第一个/第二个等）、解析失败计数与 API detail.retry_remaining；
  百度 ASR 固定 dev_pid 为普通话；SurgeryPipelineError 支持 extra 并入 HTTP detail。
- Demo：demo 路由与假 RTSP、客户端 index 与 README；BackendResolver 与配置调整。
- 可观测：消耗 TSV 日志、语音文件日志、终端 Markdown 辅助；相关测试与依赖更新。
- 注意：.env 仍被 gitignore，本地密钥不会进入本提交。

Made-with: Cursor

app/schemas.py

@@ -141,6 +141,48 @@ class SurgeryVoiceStatusResponse(BaseModel):
     )
 
 
+class SurgeryVoiceAuditItem(BaseModel):
+    """单条 `voice_confirmation_audits` 行（追溯对账用）。"""
+
+    id: int
+    confirmation_id: str
+    status: str = Field(
+        description=(
+            "recognized / rejected / parse_failed / asr_failed / invalid_audio / "
+            "upload_failed / client_stt_empty / client_stt_parse_failed 等"
+        ),
+    )
+    audio_object_key: str | None = None
+    audio_content_type: str | None = None
+    audio_size_bytes: int | None = None
+    audio_sha256: str | None = None
+    asr_text: str | None = None
+    resolved_label: str | None = None
+    options_snapshot_json: str | None = Field(
+        default=None,
+        description="当次候选项与置信度 JSON 快照。",
+    )
+    error_message: str | None = None
+    created_at: datetime = Field(
+        description="记录写入时间（UTC）。",
+    )
+
+    model_config = ConfigDict(from_attributes=True)
+
+
+class SurgeryVoiceAuditsListResponse(BaseModel):
+    """按手术号分页的语音确认审计列表。"""
+
+    surgery_id: str
+    total: int = Field(ge=0, description="该手术在表中的总条数（不受本页 limit 截断）。")
+    limit: int = Field(ge=1, le=200)
+    offset: int = Field(ge=0)
+    items: list[SurgeryVoiceAuditItem] = Field(
+        default_factory=list,
+        description="按 `created_at` 降序。",
+    )
+
+
 class PendingConfirmationOption(BaseModel):
     label: str
     confidence: float
@@ -158,6 +200,16 @@ class SurgeryPendingConfirmationResponse(BaseModel):
     created_at: datetime
 
 
+class SurgeryPendingResolveTextRequest(BaseModel):
+    """由浏览器 Web Speech 等客户端本地识别后提交的文本，语义与经百度 ASR 得到的文本相同。"""
+
+    recognized_text: str = Field(
+        min_length=1,
+        max_length=2000,
+        description="识别文本；服务端用与语音接口相同的规则解析候选项。",
+    )
+
+
 class SurgeryPendingConfirmationResolveResponse(BaseModel):
     surgery_id: str
     confirmation_id: str