Verify make_column_transformer with MCUVScaler + PLS Pipeline

[kgdunn]

Background

sklearn.compose.make_column_transformer is the standard way to apply different preprocessing to different column subsets - useful when only some X columns need centring/scaling, or when you mix MCUVScaler with OneHotEncoder on categorical features inside a Pipeline:

from sklearn.compose import make_column_transformer
from sklearn.preprocessing import OneHotEncoder
from process_improve.multivariate import MCUVScaler, PLS

ct = make_column_transformer(
    (MCUVScaler(), ["temp", "pressure", "flow"]),
    (OneHotEncoder(), ["batch_type"]),
    remainder="drop",
)
pipe = Pipeline([("ct", ct), ("pls", PLS(n_components=3))])

The sklearn Pipeline interop audit in #383 never tried this. Probable issues:

• MCUVScaler may return a pd.DataFrame while OneHotEncoder returns a sparse matrix; ColumnTransformer needs to concatenate them.
• The output of make_column_transformer is an ndarray (or sparse matrix) when mixing types - PLS.fit may need to handle that.
• get_feature_names_out (see the separate issue) becomes important here for downstream introspection.

What to do

1. Add a focused test that walks the example above end-to-end.
2. Fix anything that surfaces.
3. Document the pattern in the README if it works cleanly.

Related

Spun out from the sklearn interop audit attached to #383. Likely benefits from / requires the get_feature_names_out fix from the related issue.