start of a port of kyutai's moshi text to speech

[Codes4Fun]

https://github.com/Codes4Fun/moshi.cpp

In it's current state, doing tts, I am getting about 3 seconds of audio from 1 second of generation time on an RTX 4090, and I think there is a lot of room to improve on that.

[Codes4Fun]

I've done a binary release for those interested in demoing speech-to-speech with a hallucinating ai, minimum requirements 8GB of VRAM and RTX 2070:
https://github.com/Codes4Fun/moshi.cpp/releases/tag/v0.6.0-beta