[WIP] [NN] Refactor NN package

[jermainewang]

Description

This PR refactor the out-of-date NN package. Some guidelines:

• We won't try to unify different backend for this. User needs to explicitly import, for example dgl.nn.pytorch. This is because the interface might be quite different for different backends.
• Try to follow the convention of each backend.
• Don't forget to add docstring.

Checklist

Please feel free to remove inapplicable items for your PR.

• The PR title starts with [$CATEGORY] (such as [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

Changes

• dgl.nn.pytorch.conv.GraphConv
• dgl.nn.pytorch.softmax.EdgeSoftmax
• dgl.nn.mxnet.conv.GraphConv
• Also change the example to use these predefined modules.

Additional Note

• I've also done some small fixes on the way to modify the examples
• The incorporation of nn modules into the examples seem to have slowed down the speed of execution as there are some redundant operations of setting and popping data.

[mufeili]

@jermainewang I assume only the examples will be affected and the tutorials are left unchanged?

[jermainewang]

Yes.

[mufeili]

@jermainewang I'm done and you may take a look. There are some additional changes that I did not realize until now (the ones in sampler.h, sampler.cc, pylintrc and task_lint.sh), I suppose these are your changes?

[jermainewang]

There are some additional changes that I did not realize until now (the ones in sampler.h, sampler.cc, pylintrc and task_lint.sh), I suppose these are your changes?

Yes

[jermainewang]

Did some minor fix on MXNet side. Thanks Murphy!