Add configurable checkpoint path and validation for inference pipeline

[swagat-mishra28]

Hi,

This PR improves the robustness and usability of the inference pipeline.

Previously the checkpoint path and dataset path were hardcoded inside inference.py. This could lead to runtime errors if the checkpoint file was missing or if users wanted to run inference on a different dataset location.

This update introduces command-line arguments for the validation dataset path and checkpoint path.

Additionally, a validation step was added to ensure the checkpoint file exists before loading it, and the model is explicitly switched to evaluation mode before inference.

These changes make the inference pipeline more flexible and safer to run in different experimental setups.