Expose qnnpack's maxpool when going through aten::max_pool2d (#38896)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/38896
Current way of exposing qnnpack's maxpool2d only works if max_pool2d op is
quantized::max_pool2d. This diff moves the function about to expose it via
aten::max_pool2d when dispatch key is QuantizedCPU.
Test Plan: Quantized tests.
Reviewed By: supriyar
Differential Revision: D21690913
fbshipit-source-id: 75fb77329b915e3a3c3aac4d76359482976ca783