fix(inference): pass the token through when loading the encoder

[serhiizghama]

StaticModelPipeline.from_pretrained documents that a private hub model needs a token, and it does use it — but only for head.safetensors. The encoder load right after it calls StaticModel.from_pretrained(folder_or_repo_path) with no token, so on a private repo the head downloads fine and the model itself then fails auth. Same call in convert_legacy_pipeline, so the legacy path has it too. Passed token through in both places.

test_load_pipeline_from_hub doesn't catch this because it patches StaticModel.from_pretrained wholesale and only asserts on the head download, so I added two tests that check the token actually reaches the encoder call — one per path. Both fail on main (assert None == 'secret') and pass with the change; full suite 295 passed, ruff clean.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
model2vec/inference/model.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.

[stephantul]

Hey @serhiizghama,

Thanks for spotting this, great find!