fix: correctly set weights when w is initialized

[stephantul]

When initializing a trainable model with weights, we correctly set the weights in the initializer. However, when calling _initialize, we would call construct_weights, which would ignore the original weights. Because _initialize was always called before fitting (it is called on every fit call), this made us effectively ignore weights.

The fix is straightforward: we save the weights as _weights and initialize w to _weights if it is not None. One interesting thing is that we have to project _weights to the inverse of the sigmoid before doing so.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

Files with missing lines	Coverage Δ
model2vec/train/base.py	98.06% <100.00%> (+0.03%)	⬆️
model2vec/train/utils.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.

[Pringled]

Nice catch!