Update evaluate.py (Optimize Dice computation by removing one-hot encoding)

[dhana2403]

This PR improves the evaluate function in the segmentation pipeline by computing Dice directly on class indices instead of converting masks to one-hot format.

Changes:

Removed .one_hot() conversion for mask_true and mask_pred.

Updated Dice computation to work directly on class indices.

Abstracted repeated operations into helper functions for clarity.

Benefits:

Faster execution, especially for large images or batches.

Cleaner code with fewer tensor transformations.

No change in the resulting Dice scores, numerical results remain equivalent.

Notes:

Please verify correctness of Dice computation after these changes.