[MXNET-97] implement DepthwiseConv2dBackwardFilterKernel from tensorflow codebase

[nihui]

implement DepthwiseConv2dBackwardFilterKernel from tensorflow codebase, which is over 5x faster

Description

The current mxnet implementation calls __syncthreads() function too much, which is extemely slow.
The new code comes from tensorflow, but the variable names are adjusted for consistency.

My model uses depthwise conv heavily, and now its training time per iteration is over 5x faster on single P40 gpu. ( old 92s vs new 18s )

Checklist

Essentials

• Passed code style checking (make lint)
• Changes are complete (i.e. I finished coding on this PR)
• All changes have test coverage:
• Unit tests are added for small changes to verify correctness (e.g. adding a new operator)
• Nightly tests are added for complicated/long-running ones (e.g. changing distributed kvstore)
• Build tests will be added for build configuration changes (e.g. adding a new build option with NCCL)
• 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

[chinakook]

Tensorflow has also a 32x32 DepthwiseConv when feature maps are small. Could you add that too?

[nihui]

@chinakook hi, the special path for small maps has already been implemented in mxnet.