Bump the moshi-mlx version to 0.2.2. (kyutai-labs#240)

official 0.2.2 release

v0.2.1 after massive mess up

v2 moshi release

Lightformer implementation for MLX (kyutai-labs#215)

* Remove the step-idx.

* Support low rank embeddings.

* Use the proper number of depformer layers.

* Weights-per-step export.

* Fix the weight export script.

* Export script fix.

* Set the number of codebooks.

* Update for 16 rvq.

Bump the rust version + support for python 3.13. (kyutai-labs#205)

* Bump the rust version + support for python 3.13.

* Fix.

* Upgrade to pyo3 0.23.

* Adapt everything to the latest pyo3/numpy.

* Update the stubs.

Another CI fix attempt.

Remove arm-v7.

Bump the version of the macos runners.

Merge pull request kyutai-labs#100 from kyutai-labs/mimipyo3-max-seq-len

Increase max_seq_len for mimi-pyo3 and make it configurable.