Support for TensorRT optimization in inference pipeline

[kxgst228]

Adding TensorRT support could significantly speed up inference for NVIDIA GPUs. Should be an optional optimization step in the model loading pipeline.