[Quant] Qconv: Add qint8 support backed by xnnpack (#73669)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/73669
if built with USE_XNNPACK, support for quantized conv 1d, 2d w/ transpose and per_channel support for act dtype KQInt8 is added. For KQUInt8 act dtype qnnpack is used as before.
Reviewed By: kimishpatel
Differential Revision: D34470379
fbshipit-source-id: 23fa6c7b73627e0fd3e2fb4001f501e2bccfc2bd
(cherry picked from commit f1b05d2e6e1c8b028dab07e8338a954c8e1def98)