feat(export): Add GGUF tokenizer metadata export for llama.cpp integration

[umran666]

Description

LLaMA.cpp and local LLM runners consume tokenizers in the .gguf format (tokenizer.ggml.tokens, tokenizer.ggml.scores, tokenizer.ggml.token_type).

Scope of Work

1. Add an export method to hf_exporter.py or a new gguf_exporter.py.
2. Convert UniqToken log-probabilities and vocab tables into standard GGUF metadata key-value pairs.
3. Add a unit test verifying round-trip score extraction.

Getting Started

• Files to edit: hf_exporter.py, test_tokenizer.py