Update kfda.py

[shahedahmed2023]

Title: Update KFDA implementation to address deprecated features and numerical stability issues

Description: This PR updates the Kernel Fisher Discriminant Analysis (KFDA) implementation to address several issues:

1. Fixes deprecated features:

   • Updates OneHotEncoder usage to specify sparse_output parameter
   • Replaces scipy.sparse.eye with numpy.eye for better compatibility

2. Improves numerical stability:

   • Adds proper handling of edge cases in n_components calculation
   • Prevents division by zero in class mean calculations
   • Adds regularization to matrices for better conditioning
   • Implements custom prediction to avoid NearestCentroid warnings

3. Enhances robustness:

   • Handles broadcasting issues in mean centering
   • Adds appropriate jitter to centroids when needed
   • Updates attribute checks for better compatibility with scikit-learn

These changes maintain the same mathematical approach while making the implementation more robust across a wider range of datasets and making it compatible with newer versions of scikit-learn and scipy.

Tests have been run on various datasets to ensure the accuracy is maintained while eliminating warnings.