feat(api): use Tencent ASR flash with 16k_zh_large and dev transcript logs

Replace CreateRecTask polling with recording-file flash API, add TENCENT_APP_ID,
remove server-side pydub slicing, and log ASR recognition text at INFO in development.

Co-authored-by: Cursor <cursoragent@cursor.com>

app-expo/src/features/voice/recorder.ts

@@ -18,7 +18,7 @@ type StatusListener = (status: RecorderStatus) => void;
 type RecordingCompleteListener = (uri: string, durationMs: number) => void;
 
 /**
- * Tencent SentenceRecognition uses `EngSerViceType=16k_zh_large` and
+ * Tencent ASR flash (`16k_zh_large`); client sends ~15s m4a segments per upload.
  * `VoiceFormat=m4a`, so record speech in that shape directly instead of
  * relying on Expo's default 44.1 kHz stereo preset.
  */