feat: add POST /infer/voice endpoint for text-to-speech audio [FaaFyfxR9WAQrL7FcAgEHJvztd8cVMxvjHRS55rw1nwH]

[waterWang]

FastAPI: POST /infer/voice endpoint

Adds text-to-speech audio endpoints to the Loru API.

Changes

src/loru/api/app.py:

• Added VoiceBody Pydantic model (text + language fields)
• Added POST /infer/voice — converts text to speech and returns audio metadata
• Added POST /infer/voice/file — returns the WAV audio file directly
• Added FileResponse import for audio file streaming

tests/test_api.py:

• test_infer_voice — verifies audio metadata response
• test_infer_voice_empty_text — validates 422 on empty text
• test_infer_voice_file — verifies WAV file response with headers

API

POST /infer/voice
{"text": "Hello world", "language": "en"}

Response:
{
  "ok": true,
  "text": "Hello world",
  "audio_path": "/tmp/.../voice.wav",
  "audio_exists": true,
  "audio_size_bytes": 12345,
  "tts_backend": "OfflineStubTTS",
  "duration_sec": 1.5
}

Design

• Uses the existing get_default_tts() from loru.voice.tts — supports all TTS backends
• OfflineStubTTS generates a real playable WAV (tone, not silence)
• Both endpoints work without any external dependencies beyond the [api] extras
• POST /infer/voice/file returns a proper FileResponse with audio/wav content type

Fixes #10

Wallet: FaaFyfxR9WAQrL7FcAgEHJvztd8cVMxvjHRS55rw1nwH