Codec-bpe

- Fix for exclusive upper bound on `max_token_length` parameter to BpeTrainer. This parameter is set when passing the `--max_token_codebook_ngrams` argument in `codec_bpe.train_tokenizer`

- codec_framerate parameter should accept floats to support codecs with a non-integer framerate (e.g., Mimi, 12.5 Hz)

- Added support for Kyutai Lab's [Mimi codec](https://huggingface.co/kyutai/mimi), an amazing new codec with a 12.5 Hz framerate! Simply add `--use_mimi` when encoding audio with `codec_bpe.audio_to_codes` to encode using the Mimi model.
**Note:** Until Mimi is included in a stable release of HuggingFace Transformers, you need to install Transformers from source:
bash
pip install git+https://github.com/huggingface/transformers.gitmain

- Fix default output path in the `codec_bpe.audio_to_codes` tool

First release of codec-bpe!

- Installable using `pip`
- All tools runnable as submodules, e.g., `python -m codec_bpe.train_tokenizer ...`
- Improved usage examples