Add data preprocessing utilities for image normalization and augmentation

[kamalahasiniburra]

Summary

This PR adds a reusable data preprocessing utilities module for the DeepLense project with comprehensive tests.

What is included

preprocessing_utils.py

• normalize_images(): Image normalization with minmax, zscore, and robust methods
• • augment_image(): Augmentation transforms including flip, rotate, and Gaussian noise
• • stratified_split(): Stratified train/val/test splitting to maintain class balance
• • compute_class_weights(): Class weight computation for imbalanced datasets
• • create_image_patches(): Extract overlapping patches for patch-based training

test_preprocessing_utils.py

• 12 comprehensive tests covering all preprocessing functions
• • Tests for normalization correctness, augmentation behavior, split validity, and edge cases

Why this is useful

Data preprocessing is fundamental to every DeepLense sub-project but is often re-implemented. This shared module standardizes preprocessing across sub-projects, ensuring consistent data handling and reproducibility. All 12 tests pass locally.