Download tokenizer artifacts in CI instead of storing them in `tests/testdata/models`

[khaledsulayman]

It was raised in discussion on #364 that we shouldn't be storing tokenizer artifacts in our testdata.

Points in favor

• artifacts may become out-of-date overtime, leading to inaccurate tests
• we should avoid adding an unnecessary amount of data to our codebase

Proposed solution

We download the necessary artifacts during CI runs in order to run tests