Added HRNet18 click based interactive segmentation (#3729)

[xiong-jie-y]

The commit adds serverless function that has a click based
interactive segmentation model(HRNet18) that performs better than
f-BRS method statistically.

Motivation and context

I was trying to do hand segmentation and my task was a bit tough with f-BRS model, because a segment sometimes doesn't converge to the intended segment.
This HRNet model seems more intuitive, and also statistically perform better from the paper.

How has this been tested?

This feature is manually tested by following procedure.

1. deploy

cd serverless
nuctl deploy --project-name cvat --path pytorch/saic-vul/hrnet/nuclio --file pytorch/saic-vul/hrnet/nuclio/function-gpu.yaml

2. test HRNet interactor in cvat (HRNet will be shown as a interactor by deployment)

Checklist

• I submit my changes into the develop branch
• I have added description of my changes into CHANGELOG file
• I have updated the documentation accordingly
• I have added tests to cover my changes
• I have linked related issues (read github docs)
• I have increased versions of npm packages if it is necessary (cvat-canvas,
  cvat-core, cvat-data and cvat-ui)

License

• I submit my code changes under the same MIT License that covers the project.
  Feel free to contact the maintainers if that's a concern.
• I have updated the license header for each file (see an example below)

# Copyright (C) 2021 Intel Corporation
#
# SPDX-License-Identifier: MIT

[nmanovic]

@xiong-jie-y , thanks for the great contribution!

[xiong-jie-y]

@nmanovic Thank you for reviewing and merging!