[JIT] Separate GPU implementation of frozen_conv_add_relu_fusion.cpp (#68149)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/68149
JIT optimization passes are part of the CPU-only build (i.e. necessary GPU flags are not passed in). This separates the implementation of frozen_conv_add_relu_fusion so that the GPU-enabled implementation is registered at runtime (if it is available)
Test Plan: Imported from OSS
Reviewed By: H-Huang
Differential Revision: D32773666
Pulled By: davidberard98
fbshipit-source-id: c83dbb88804bdef23dc60a6299acbfa76d5c1495
