Add input validation in predict() for feature shape and NaN handling

[2024itb047samata]

This PR improves the robustness of the predict() method by adding input validation for feature shape and NaN values.

• Ensures input X is converted to a 2D NumPy array
• Automatically reshapes 1D input to (n_samples, 1)
• Validates that the number of features matches the trained model (n_features_)
• Raises a clear error if input contains NaN values

Previously, predict() could silently accept invalid inputs (e.g., incorrect shape or NaN values), leading to unexpected behavior or errors deeper in the pipeline. This change aligns the behavior with standard practices in libraries like scikit-learn and improves usability.