Add adaptive_avgpool2d to the set of fusible ops (#56180)
Summary:
This improves mobilenetv3 perf from ~1300msto 1147ms (~12%)
Pull Request resolved: https://github.com/pytorch/pytorch/pull/56180
Reviewed By: Chillee
Differential Revision: D27840860
Pulled By: Krovatkin
fbshipit-source-id: 6ce38e93fd2f55e68a69c34b45271743f84a13b8
