Gemma3 initial commit

[chapman20j]

Resolves #100
Adds the Gemma3 model to the Bonsai Repo. This first commit is just a working version, but I am still working on optimizing it.

Reference
Refer to Issue #100

Checklist

• I have read the Contribution Guidelines and used pre-commit hooks to format this commit.
• I have added all the necessary unit tests for my change. (run_model.py for model usage, test_outputs.py and/or model_validation_colab.ipynb for quality).
• (If using an LLM) I have carefully reviewed and removed all superfluous comments or unneeded, commented-out code. Only necessary and functional code remains.
• I have signed the Contributor License Agreement (CLA).

[jenriver]

Also, please make sure your selective tests are passing

[jenriver]

Currently this is a global KV cache implementation and doesn't introduce kv cache mem reduction benefits in gemma 3 -

Could we have something like this to account for local vs. global?

def init_cache(...) -> Cache:
    full_cache_size = 2 ** math.ceil(math.log2(max(token_len + generate_steps, 1)))
    window_size = cfg.text_config.sliding_window # Typically 1024
    
    caches = []
    for i, layer_type in enumerate(cfg.text_config.layer_types):
        size = full_cache_size if layer_type == AttentionMode.FULL else window_size
        caches.append(LayerCache(cfg.text_config, batch_size, size, dtype))
    return caches

We can also fix the k / v cache update parts to be based on window_size modulo.