Optionally remove the REGEXP from ByteLevel pre_tokenizer

[Narsil]

This is sort of done here:

https://github.com/huggingface/tokenizers/pull/896/files#diff-c2be5839194fbc7e052d4600684eeb35f0db504b5b4df6565f01057762e23774R46

There might be some changes which might make it better, like using None instead of "no_regexp" in Python.
(In RUST, since the regexp is hardcoded, and we definitely don't want to enable users to choose, since it should be done as another pre_tokenizer if you want custom regexp, going with the simple Enum instead of Option seems more correct as a first approach.