Add support for model quantization.

Is your feature request related to a problem? Please describe.
In order to get a better inference performance, there are ways to quantize a deep learning model to a lower precision model like int8/int4 with an acceptable precsion decrease. Here are some examples:

1. Pytorch Official
2. NVIDIA library
3. ONNX library

Medical images always cost plenty of inference time because of the 3D shape and large size. Since MONAI has already supported the onnx and trt export, it would be better to leverage the quantization feature supported by these formats and get a better latency for the medical image inference. What's more this will benefit the edge and network applications, both of which would be benefit from the low latency.

Describe the solution you'd like
APIs to convert, save, load and deploy quantization models.
Functions to perform the corresponding actions in python scripts.

Start by reviewing MONAI’s existing ONNX and TensorRT export support and the linked PyTorch, NVIDIA, and ONNX quantization documentation. Define Python APIs for converting, saving, loading, and deploying quantized models, with completion demonstrated by working quantization workflows for medical-image inference.