Pass sklearn check_estimator test suite

[jc-macdonald]

Summary

Add test_sklearn_compat.py that runs sklearn's check_estimator on all public estimator/transformer classes and passes.

Details

This is the acceptance test for full sklearn compatibility. It verifies that:

• VBPCA passes all applicable check_estimator checks
• MissingAwareStandardScaler, MissingAwareMinMaxScaler, MissingAwareOneHotEncoder, AutoEncoder all pass
• VBPCApy works in sklearn.pipeline.Pipeline
• GridSearchCV can tune VBPCA hyperparameters
• clone() correctly round-trips all parameters

Test file structure

from sklearn.utils.estimator_checks import parametrize_with_checks

@parametrize_with_checks([VBPCA(n_components=2)])
def test_sklearn_compatible(estimator, check):
    check(estimator)

Plus integration tests:

• Pipeline: Pipeline([('scaler', MissingAwareStandardScaler()), ('pca', VBPCA(n_components=2))]) fit/transform
• GridSearchCV: sweep n_components with cross-validation
• clone: verify parameter fidelity

Depends on

• Issue: Inherit from sklearn BaseEstimator + TransformerMixin
• Issue: Add sklearn_tags declarations
• Issue: Input validation at the public API boundary
• Implement transform(new_data) for out-of-sample projection #18 (out-of-sample transform — needed for Pipeline transform on test fold)

Acceptance criteria

• pytest tests/test_sklearn_compat.py passes with zero failures
• No xfail or skip markers on check_estimator tests