Optimizing the tokenizer performance

[PicoCreator]

The current implementation is sub-optimal, and easily 10x slower then the HF rust implementation.
Any improvements here would be appreciated.

[cahya-wirawan]

Hi @PicoCreator
I just found out that you already created RWKV Tokenizer for Nodejs. I created also RWKV Tokenizer for node, but it is a WebAssembly version based on my Rust RWKV Tokenizer implementation. The performance should be near the native Rust version, which is currently the fastest Tokenizer :-)