FIX: Add scikit-learn >= 1.6.0 compatibility for _validate_data

[J4nvg]

Description

The Issue

Users installing imbalanced-ensemble alongside scikit-learn >= 1.6.0 encounter an AttributeError when calling .fit() on ensemble classifiers (e.g., SelfPacedEnsembleClassifier).

AttributeError: 'SelfPacedEnsembleClassifier' object has no attribute '_validate_data'

This happens because scikit-learn version 1.6.0 removed the private _validate_data method from the BaseEstimator class and moved it to a public standalone utility function (sklearn.utils.validation.validate_data).

The Solution

This PR introduces a backward-compatible bridge in imbens/ensemble/base.py to handle the version difference dynamically without breaking older setups.

• Added a try/except block to safely import validate_data from sklearn.utils.validation if it exists.
• Overrode _validate_data in BaseImbalancedEnsemble to act as a routing method:
  • If scikit-learn >= 1.6.0, it passes the arguments to the new utility function.
  • If scikit-learn < 1.6.0, it falls back to super()._validate_data().

Testing

• Successfully built locally.
• Verified that the AttributeError is resolved when running with scikit-learn==1.6.0.
• Ran the local pytest suite. All core ensemble tests pass successfully, confirming that the fallback preserves behavior for existing logic.