[DOC] Add warning about over-sampling limitations

[Asif-0209]

Reference Issue

Fixes #1101

What does this implement/fix? Explain your changes.

This PR adds a visible warning at the top of the over-sampling user guide to set expectations about over-sampling methods (e.g., random over-sampling, SMOTE, ADASYN). The warning notes that these methods may not improve ranking metrics (e.g., ROC-AUC) and can degrade probability calibration, and points users to common alternatives such as class_weight, sample_weight, decision threshold tuning, or collecting more minority data.

Any other comments?

This change is documentation only.

[Asif-0209]

Hi, I opened a PR adding a visible warning in the over-sampling documentation about limitations regarding ranking metrics and probability calibration, and pointing to common alternatives. PR: #1160