Integrate cache changes

[dleemiller]

No description provided.

[dleemiller]

• Removed weights_dir option in load()
  • Replaced with cache_dir
  • added path resolution for strings
  • unified tokenizer/weights cache to same dir

          Weights:
            1. Check in project root (`project_root / wordllama / weights / ...`).
            2. If not found, check in `cache_dir / weights / ...`.
            3. If still not found, download from Hugging Face to `cache_dir / weights / ...`.
        Tokenizer:
            1. Check in project root (`project_root / wordllama / tokenizers / ...`).
            2. If not found, check in `cache_dir / tokenizers / ...`.
            3. If still not found, download from Hugging Face to `cache_dir / tokenizers / ...`.   
  

Example Usage:

wl = WordLlama.load(dim=64, cache_dir="some_different_cache_folder/")