feat(speech): add Tencent Cloud ASR and TTS backend

[kaileliu]

This PR adds a Tencent Cloud speech backend for services/speech, allowing the existing liaison voice pipeline to use cloud ASR and TTS by setting SPEECH_BACKEND=tencent while keeping the liaison-facing contracts unchanged (robonix/service/speech/asr_stream and robonix/service/speech/tts). It adds Tencent real-time ASR over WebSocket, Tencent TextToVoice TTS, env-based credential loading, and an adapter that converts Tencent ASR cumulative partial results into the incremental ASR events expected by liaison. Validation included Python syntax checks and an end-to-end liaison demo using input.wav -> mock mic -> Tencent ASR -> liaison -> mock pilot -> Tencent TTS -> mock speaker -> output.wav; the ASR transcript was 请介绍一下你自己。, the output WAV was generated as 16 kHz mono pcm_s16le, and the measured latency from ASR_FINAL to TTS_DONE was 0.653s, meeting the target of less than 3 seconds.