NNC depthwise conv2d implementation (#54920)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/54920
Add a depthwise convolution implementation and reasonably good
schedules for 3x3 stride=1,2.
ghstack-source-id: 126076113
Test Plan: new tensorexpr test: Conv.DepthwiseConv2D
Reviewed By: ZolotukhinM
Differential Revision: D27413745
fbshipit-source-id: 833da6072b655fbe2b679704e9d56a08e1bf7e7e
