[Feature] Add multi-GPU UnifiedTensor unit test

[davidmin7]

Description

This is a followup PR of #3086, which adds a multi-gpu unit test case for the PyTorch backend using UnifiedTensor.

Checklist

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [$CATEGORY] (such as [NN], [Model], [Doc], [Feature]])
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage
• Code is well-documented
• To the my best knowledge, examples are either not affected by this change,
  or have been fixed to be compatible with this change
• Related issue is referred in this PR
• If the PR is for a new model/paper, I've updated the example index here.

Changes

• Add multi-gpu unified tensor test.

[dgl-bot]

To trigger regression tests:

• @dgl-bot run [instance-type] [which tests] [compare-with-branch];
  For example: @dgl-bot run g4dn.4xlarge all dmlc/master or @dgl-bot run c5.9xlarge kernel,api dmlc/master

[VoVAllen]

Bus error is probably related to the docker settings. Will take a look at this tommorow

[davidmin7]

Bus error is probably related to the docker settings. Will take a look at this tommorow

@VoVAllen Thanks for the quick reply! Based on my quick research, I believe the docker's shared memory space is too small. Here is the relevant link: pytorch/pytorch#2244.

[classicsong]

Bus error is probably related to the docker settings. Will take a look at this tommorow

Do we extend the shm size for the docker container?

[VoVAllen]

Fixed shared memory problem

[classicsong]

LGTM