[akshitkhokhani]

FeatureSelector.fit gives ndarray targets a new default index even when the feature DataFrame has a different index. This breaks sklearn pipeline cross-validation with an index assertion. If the DataFrame instead has permuted integer index labels, selection can silently associate targets with the wrong rows and discard a predictive feature.

Construct the target Series with the feature DataFrame's index so ndarray targets follow row order. Explicit Series targets keep their existing label alignment. The fit documentation clarifies both behaviors.

Adds regressions for nondefault and permuted indexes, reordered Series targets, and an actual sklearn pipeline cross-validation workflow. Related to the pipeline workflow discussed in #960; that issue does not establish this specific ndarray cause.

Validation: all 10 selector unit tests pass, including existing ndarray-input and multiclass cases. Both new tests fail before the fix. Black 22.12.0, isort 5.12.0 and git diff --check pass. No full-suite claim.