TensorRT Export for NITEC Model

This PR adds TensorRT support for the pretrained NITEC model.

Implemented Features

• ONNX export (opset 17, dynamic batch size)
• TensorRT engine build script (FP32 + optional FP16)
• TensorRT inference demo (webcam + image-based for Colab)
• PyTorch vs TensorRT validation (np.allclose comparison)
• README documentation for full TensorRT workflow

TensorRT Version

Tested with: TensorRT 10.15.1.29

Validation

np.allclose: True
Max abs err: ~2.98e-07

Known Limitations

• TensorRT requires NVIDIA GPU + CUDA.
• Webcam demo does not run in Google Colab.
• ONNX export required legacy exporter (dynamo=False) for TensorRT parser compatibility.