Refactoring test.py to support multiple dataset paths and modular transforms

[BeathovenGala]

What this PR does

Refactors DeepLense_Diffusion_Rishi/utils/test.py into a robust, configurable script for visualizing lens .npy data. The original script loaded data from a single hardcoded path, used a fixed index, and had minimal error handling. This PR:

• Supports multiple dataset paths (e.g., cdm_regress_multi_param, npy_lenses, real_lenses_dataset).
• Makes the transform pipeline modular via get_transforms().
• Handles missing files, invalid directories, and constant‑value arrays gracefully.
• Improves maintainability and reproducibility for the DeepLense Diffusion project.

Changes for Issue #124

• Use SCRIPT_DIR and DEFAULT_DATA_PATHS to locate datasets relative to the script location:

  DEFAULT_DATA_PATHS = [
      os.path.join(SCRIPT_DIR, "../Data/cdm_regress_multi_param_model_ii/cdm_regress_multi_param/"),
      os.path.join(SCRIPT_DIR, "../Data/npy_lenses-20240731T044737Z-001/npy_lenses/"),
      os.path.join(SCRIPT_DIR, "../Data/real_lenses_dataset/lenses"),
  ]

• Replaced hardcoded logic with argparse arguments (--data_dirs, --index, --output_dir, --filename).

• Print selected directory, file path, original shape/range, and transformed tensor shape/range for better debugging.