Add reusable model evaluation utilities with comprehensive tests

[kamalahasiniburra]

Summary

This PR adds a reusable evaluation utilities module (evaluation_utils.py) and a comprehensive test suite (test_evaluation_utils.py) for the DeepLense project.

What's included

evaluation_utils.py

• compute_classification_metrics() : Computes accuracy, per-class precision/recall/F1, and macro-averaged metrics for gravitational lens classification
• • compute_confusion_matrix() : Builds confusion matrix without requiring sklearn dependency
• • compute_regression_metrics() : Computes MSE, RMSE, MAE, and R-squared for lens parameter estimation tasks
• • compute_roc_auc_ovr() : One-vs-rest ROC AUC computation for multi-class classification
• • format_metrics_table() : Formats metrics dictionary as a readable table string

test_evaluation_utils.py

• 8 comprehensive tests covering all functions including edge cases
• • Tests for perfect predictions, partial misclassifications, confusion matrix correctness, regression accuracy, and input validation

Why this is useful

Currently, each DeepLense sub-project implements its own evaluation logic. This shared module provides standardized metrics computation that can be imported across different sub-projects, ensuring consistent evaluation and reducing code duplication.

All 8 tests pass locally.