Add ModelOpt support

[sam-india-007]

Change in builder utility to enable an option to use NVIDIA's TensorRT Model Optimizer https://github.com/NVIDIA/TensorRT-Model-Optimizer

[sam-india-007]

@microsoft-github-policy-service agree company="NVIDIA"

[kunal-vaishnavi]

Thank you for your contribution!

How long does the int4_awq_clip quantization take with modelopt? The goal of the model builder is to quickly produce and save the final ONNX model to disk within a few minutes max. If int4_awq_clip takes a while, a note will need to be added somewhere so users are aware in advance that the model builder will take time to run with int4_awq_clip quantization enabled.

[sam-india-007]

ModelOpt quantization with int4_awq_clip can take significant amount of time, but results in smaller and better quantization. Added a disclaimer in the README to this end