[codex] Add local MLX and KittenTTS voice backends

[cameronbergh]

What changed

• add a local kitten TTS provider for Hermes
• add a local/self-hosted mlx TTS provider that can call OpenClaw-compatible MLX audio servers
• add KittenTTS and MLX Audio Server to the Hermes TTS setup menu and config docs
• accept JSON {\"text\": ...} responses from OpenAI-compatible STT servers when Hermes requests text transcription

Why

This makes it easier to run Hermes fully against local audio infrastructure on Apple Silicon, including self-hosted MLX speech services and lightweight local TTS options.

User impact

• Hermes can now use a self-hosted MLX speech server for TTS voices such as Samantha and Judy
• Hermes can now use KittenTTS as a local voice backend
• OpenAI-compatible local STT servers that return JSON text payloads work more reliably with Hermes

Validation

• python3 -m py_compile tools/transcription_tools.py tools/tts_tool.py tools/kitten_tts_synth.py
• local smoke tests against a running MLX server for STT and TTS

[teknium1]

Closing as superseded by #13395 — #13395

Both this PR and @AxDSan's #2109 proposed KittenTTS support. #2109 was submitted first (March 20 vs March 28), so we salvaged his implementation onto current main with the full integration layer (setup wizard, tools picker, auto-install, docs, tests) added on top. Your MLX voice backend work is a separate contribution — feel free to reopen a focused PR for just that piece and we'll review.

Thanks for the contribution!