fix: converter hf now handles byte characters. Closes #188

[antoine-sac]

The converter-tokenizer-hf is now aware of byte characters (such as "<0x0A>" and parses them correctly as the actual character (such as a newline "\n".

This is useful for mistral, tinyllama, and others using the same tokenizing method.

See ggml-org/llama.cpp#4622 for more context.

Fix #188.

[b4rtaz]

Thanks @antoine-sac!