Warn when imbalance handling adds validation samples

[kiwoongyoon]

Title: Warn when imbalance handling adds validation samples

Summary

• Emit a UserWarning when _handle_drastic_imbalance appends duplicated samples before validation.
• Include the number of appended samples and note that validation folds and out-of-fold predictions can include duplicated rows.
• Extend the focused imbalance tests to cover the new warning for direct imbalance handling and AutoML.fit.

Test evidence

• conda run -n mljar python -m pytest tests/tests_automl/test_handle_imbalance.py -q passed: 3 passed in 13.87s.
• git diff --check passed.
• conda run -n mljar python -m py_compile supervised/base_automl.py tests/tests_automl/test_handle_imbalance.py passed.

Risks or notes for maintainers

• The existing upsampling behavior is unchanged; this patch only makes it visible when it happens.
• The warning uses Python warnings.warn. If preferred, this can be adapted to the project logger in a follow-up.