Add scikit-learn set_output() API support to cuml.accel

[csadorf]

Currently, cuml.accel does not support scikit-learn's set_output() API, which was introduced in scikit-learn 1.2.0 to control the output format of transformers and estimators. This feature allows users to specify whether they want outputs as numpy arrays, pandas DataFrames, or other formats.

Expected Behavior

All cuml.accel estimators that implement transform() or predict() methods should support the set_output() API. This includes:

• PCA
• TruncatedSVD
• KNeighborsClassifier
• KNeighborsRegressor
• NearestNeighbors
• And other relevant estimators

Dependencies

• scikit-learn >= 1.2.0 (for set_output API)
• pandas (for DataFrame output support)

Related Issues

• Document that cuml.accel does not support the sklearn set_output() API among known limitations #6605

Acceptance Criteria

• All relevant estimators support set_output()
• Tests pass for both numpy and pandas output formats
• Documentation is updated to reflect the new functionality
• No regression in existing functionality