fix: divide by length 1 instead of 1e-16

[stephantul]

Mean pooling divided by the number of tokens plus 1e-16. For a text without any tokens, the backward pass divided the gradient by 1e-16 after normalization. This overflows to inf and turned into NaN in the token weight gradient. Clamp the length to at least 1 instead: non-empty texts are unaffected, and empty texts get a zero embedding and a zero gradient.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
model2vec/train/base.py	99.54% <100.00%> (ø)

🚀 New features to boost your workflow:

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

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes how the model handles empty token sequences.

The PR appears safe to merge; no actionable regression was identified.

_{Reviews (1) · Last reviewed commit: "fix(train): avoid NaN gradients for text..."}