Add suport for CSM

[gtokman]

Sesame Labs open sourced a conversational speech model on HuggingFace. A developer has ported it to MLX: https://github.com/senstella/csm-mlx.

It would be amazing to add it to the Swift examples so developers can start expirementing with audio generation from speech.

[DePasqualeOrg]

I tested out the Python MLX repo yesterday to get familiar with it. Opening a draft PR would be even more useful than an issue, since in general we're all aware of these new developments, and the limiting factor is people's willingness and ability to work on something new.

[LaurentMazare]

In case anyone gives this a stab, note that csm uses the mimi codec and we've released an mlx-swift implementation of mimi in the moshi-swift repo so hopefully that part should be covered (the python implementation re-uses our moshi_mlx package for mimi).