Add vocabulary type for token-free models that work on raw bytes

[uwu-420]

I think LLMs that directly work on raw bytes will become more interesting in the near future. Some interesting work in that direction:

• ByT5: Towards a token-free future with pre-trained byte-to-byte models
• Bytes Are All You Need: Transformers Operating Directly On File Bytes
• MambaByte: Token-free Selective State Space Model

I think it would be useful if something like LLAMA_VOCAB_TYPE_RAW_BYTES would be added to enum llama_vocab_type but I don't know what kind of changes that would imply elsewhere. That kind of vocabulary would still require special tokens of course.

[uwu-420]

Opened a feature request issue for visibility: #7763