Add MXNet GPU + GluonNLP and GluonCV toolkits

[Ishitori]

This pull request adds GPU version of MXNet to GPU kernel + GluonNLP and GluonCV toolkits to both kernels.

I use pure (not MKL version) of MXNet in case MKL is not supported on the host.
I use CUDA 9.2 version, as I see other packages of GPU kernel relies on 9.2 version of CUDA. Feel free to change to MKL versions if it is guaranteed to present on the host: mxnet-mkl and mxnet-cu92mkl packages for CPU and GPU respectively.

Relates to #452
@rosbo, thanks for fixing the master!

[rosbo]

Hi @Ishitori,

Can you add smoke tests for GluonNLP + CV and also tests that exercises the GPU?

See the tensorflow tests for an example:
https://github.com/Kaggle/docker-python/blob/master/tests/test_tensorflow.py

For tests that requires a GPU, add the @gpu_test annotation. See example linked above.

Thank you

[rosbo]

Thank you @Ishitori for adding the tests.

[rosbo]

Hi @Ishitori,

We have released new docker python images yesterday. Your new packages are available on Kaggle Kernels :)

Thank you