LiveKit WebRTC end-to-end validation with real audio

## Summary

LiveKit token API and worker profile exist. Roadmap still wants e2e with real audio, not just token 503/404 tests.

## Context

- `tests/integration/test_livekit_token_api.py`
- `tests/unit/test_livekit.py` (skipif)
- `src/voxforge/modules/livekit_gateway/`
- `docker-compose.prod.yml` profile `livekit`

## Scope

- [ ] Connect a client, publish audio, get a pipeline turn, confirm TTS or transcript
- [ ] Document required LiveKit Cloud or self-hosted server
- [ ] Prefer nightly/manual over PR CI unless fully mockable

## Acceptance criteria

A repeatable script or test that fails loudly when LiveKit is misconfigured, and is skipped when unset.

Start with tests/integration/test_livekit_token_api.py, tests/unit/test_livekit.py, and src/voxforge/modules/livekit_gateway/ to understand the existing LiveKit paths, then inspect the livekit profile in docker-compose.prod.yml. The finished work should provide a repeatable real-audio script or test, document the required server setup, fail clearly when misconfigured, and skip when configuration is absent.

Written by the indexing model from the issue text.