Compute and print mean single-pass uncertainty metrics during model test

[samrat-rm]

This change adds single-pass entropy-based uncertainty estimation to the model testing pipeline.
The model now computes uncertainty from softmax probabilities and reports the mean uncertainty alongside Pixel Accuracy and mIoU in the terminal output.

This implementation currently serves as a prototype for uncertainty estimation during model evaluation and can also be extended to the model training pipeline in the future for deeper analysis and experimentation.

Changes

• Compute entropy-based uncertainty from model logits.
• Aggregate mean uncertainty across the test dataset.
• Print mean uncertainty together with existing evaluation metrics.

AI usage disclosure :
AI tools like Chat GPT were used to assist with understanding concepts and drafting parts of the implementation. The final code was reviewed, tested, and validated.