[FEATURE] Add ONNX quantization export

[alphinside]

Describe the solution you'd like
Add capability to quantize ONNX model to another precision e.g. fp16