[Pytorch][ferraris] Move native/utils/Factory.cpp to aten_cpu lib (#81356)
Summary:
Stacked conv2d uses empty_with_tail_padding from Factory.cpp. This was
available only in aten_native_cpu but linking directly against that is not
wise since aten_native_cpu contains references to other ops that are part of
aten_cpu and selective build.
Thus for the purpose of padded tensor allocation I decided to bring this out in
aten_cpu.
Test Plan:
CI
buck build
fbsource//xplat/caffe2/fb/custom_ops/ferraris:stacked_conv2dAndroid#arm-64,shared
Reviewed By: cccclai
Differential Revision: D37787231
Pull Request resolved: https://github.com/pytorch/pytorch/pull/81356
Approved by: https://github.com/cccclai
