Allow overriding the gated Llama-3 tokenizer used for the LLaMa3 prompt style - #151

Summary

format_prompt_generation renders the LMStyle.LLaMa3 prompt by loading the tokenizer of meta-llama/Meta-Llama-3-8B-Instruct, purely to call apply_chat_template. That repo is gated, so anyone whose HF account has not been granted access to it gets an HTTP 403 from AutoTokenizer.from_pretrained and every Llama-3-style evaluation fails, even when the model being evaluated is a different (accessible) Llama 3 checkpoint.

This PR keeps the default unchanged and adds an environment-variable override, LCB_LLAMA3_TOKENIZER, so the chat template can be rendered from any Llama 3 family checkpoint the user can access (the chat template is the same across Llama 3 / 3.1 / 3.2 instruct models), e.g.

LCB_LLAMA3_TOKENIZER=meta-llama/Llama-3.2-3B-Instruct python -m lcb_runner.runner.main --model ... --scenario codegeneration

No behaviour change when the variable is not set.