feat(tts): add Typecast TTS extension

[hmmhmmhm]

Summary

• Add a Typecast TTS extension backed by the official typecast-python SDK.
• Stream Typecast WAV output, strip the initial WAV header, and forward PCM16 mono audio to TEN at Typecast's fixed 32 kHz sample rate.
• Expose the API endpoint as params.url, with https://api.typecast.ai as the production default.
• Add runtime, configuration, vendor-error, cleanup, PCM conversion, and TTS guarder coverage.

Validation

• Typecast TEN runtime tests: 9 passed
• TTS guarder full suite: 15 passed, 2 skipped (HTTP TTS does not emit WebSocket connection status; subtitle alignment is opt-in)
• GitHub Actions workflow reproduced locally with ghcr.io/ten-framework/ten_agent_build:0.7.14:
  • voice-assistant: check, build-agent-deps, lint, all extension tests, and Go server tests passed
  • voice-assistant-realtime: check, build-agent-deps, lint, all extension tests, and Go server tests passed
  • Black: 860 files unchanged
  • pylint: 10.00/10
• python3 -m compileall, JSON validation, and git diff --check passed

TTS Guarder Snapshot

================== 15 passed, 2 skipped in 363.11s (0:06:03) ==================

Remaining Manual Check

• TEN-Agent webpage audio-session verification requires valid Agora and STT credentials. The local workspace contains the Typecast credential but not those required credentials, so this check is still pending.

[wangyimin-agora]

1. Please add more test cases. You can refer to other TTS extensions, for example, rime_ttp_tts. Moreover, please ensure all test cases pass in CI.
2. Please run tts guarder for your new tts extension and ensure all cases can pass, then add a snapshot in the PR description or comment.
3. Please verify your TTS extension on the TEN-Agent webpage, which provides AI communication.

[hmmhmmhm]

Addressed the review items in 13a7779 and bf1e7dc:

• Renamed the public endpoint setting from host to url.
• Expanded the Typecast runtime suite to 9 cases covering config normalization/validation, empty input, vendor errors, WAV-to-PCM conversion, successful streaming, and SDK cleanup.
• Added all TTS guarder property fixtures and fixed-rate handling.
• TTS guarder result: 15 passed, 2 skipped in 363.11s.
• Reproduced the CI workflow locally in the official build image for both voice-assistant and voice-assistant-realtime; Black, dependency build, pylint, all extension tests, and Go server tests passed.

The two guarder skips are expected for an HTTP TTS extension: WebSocket connection-status reporting and opt-in subtitle alignment. TEN-Agent webpage audio verification remains pending because the local workspace does not contain the required Agora/STT credentials.

[hmmhmmhm]

Rebased the validation against the latest upstream state by merging origin/main in 2e5d539. The only conflict was the fixed-sample-rate guarder list; both the newly added EZ-AI provider and Typecast are retained. After resolving it, the Typecast guarder was rerun from a clean install and passed again: 15 passed, 2 skipped in 362.91s.

[YiminW]

Hi, if TEN-Agent webpage audio verification cannot be finished in the coming few days, please provide pcm dump generated by TTS Guarder tests. I will check the audio file.