feat: Integration or regression tests

[stephantul]

This PR adds a couple of integration/regression test baselines. These are mostly meant to guard against unforeseen changes in distillation. They are not run in CI, because they download full models. I've added three models to start:

1. minilm
2. bge-base
3. embeddinggemma
4. e5-multilingual
5. gte-modernbert

These are added as defaults. You can add new models by running make test_integration_update. You should only run this after changing the code.

Comparing the results without regenerating is done by running make test_integration. This will fail when the new results deviate significantly from the stored ones.

The integration tests surfaced a bug with BPE models and added vocabulary. An edge case, also fixed in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
model2vec/tokenizer/tokenizer.py	100.00% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.