Move tokenizer vocab `.gzip` into `sam3`?

Thanks for the amazing contribution!

Any chance we can move the [tokenizer vocab](https://github.com/facebookresearch/sam3/blob/main/assets/bpe_simple_vocab_16e6.txt.gz) file into the `sam3` module itself? I think this is general best practice since it's an intrinsic part of the model?

Reason being is because if one wants to add `sam3` as a submodule or `uv add git+https://github.com/facebookresearch/sam3.git` it will only install the `sam3` module and when loading `sam3` with defaults it searches for `../assets/` which is not included, thus raising an error.

Start with assets/bpe_simple_vocab_16e6.txt.gz and the sam3 default-loading entry point that currently searches ../assets/. Check how uv add git+https://github.com/facebookresearch/sam3.git packages the module and verify that the tokenizer vocabulary is included and defaults load successfully from the installed sam3 module.

Written by the indexing model from the issue text.