tests : improve UGM tokenizer test coverage

[CISC]

Adds nomic-bert-moe vocab for improved test coverage of UGM tokenizer.

[ggerganov]

We should avoid adding more vocab files because they increase the size of the repository significantly. Ideally, the tests should download these files from HF and we can avoid source-controlling them here.

We can merge this one for now.

[CISC]

In that case it would make sense to have a vocab repo in ggml-org on HF?

[ggerganov]

Yes. If you (or anyone else) is interested in implementing this, I will invite you to the HF org so you can set the repo. LMK

[CISC]

Sure, I can do that, I sent a request.

[CISC]

@ngxson I need the common_download_file_multiple function to be able to download and cache all the files in a repo, any thoughts around exposing this (and common_download_file_single I guess)?