Update Custom C++ and CUDA Operators to use PyTorch stable API/ABI #5621

Job	Run time
build / pytorch_tutorial_build_worker (7, 15, linux.g5.4xlarge.nvidia.gpu)	28m 24s
build / pytorch_tutorial_build_worker (5, 15, linux.g5.4xlarge.nvidia.gpu)	24m 55s
build / pytorch_tutorial_build_worker (8, 15, linux.g5.4xlarge.nvidia.gpu)	22m 58s
build / pytorch_tutorial_build_worker (4, 15, linux.g5.4xlarge.nvidia.gpu)	38m 0s
build / pytorch_tutorial_build_worker (1, 15, linux.g5.12xlarge.nvidia.gpu)	17m 7s
build / pytorch_tutorial_build_worker (10, 15, linux.g5.4xlarge.nvidia.gpu)	26m 57s
build / pytorch_tutorial_build_worker (12, 15, linux.g5.4xlarge.nvidia.gpu)	17m 50s
build / pytorch_tutorial_build_worker (13, 15, linux.g5.4xlarge.nvidia.gpu)	20m 45s
build / pytorch_tutorial_build_worker (2, 15, linux.g5.4xlarge.nvidia.gpu)	29m 39s
build / pytorch_tutorial_build_worker (14, 15, linux.g5.4xlarge.nvidia.gpu)	17m 56s
build / pytorch_tutorial_build_worker (3, 15, linux.g5.4xlarge.nvidia.gpu)	26m 50s
build / pytorch_tutorial_build_worker (11, 15, linux.g5.4xlarge.nvidia.gpu)	28m 56s
build / pytorch_tutorial_build_worker (15, 15, linux.g5.4xlarge.nvidia.gpu)	16m 59s
build / pytorch_tutorial_build_worker (9, 15, linux.g5.4xlarge.nvidia.gpu)	23m 6s
build / pytorch_tutorial_build_worker (6, 15, linux.g5.4xlarge.nvidia.gpu)	30m 14s
build / pytorch_tutorial_build_manager	23m 5s
	6h 33m 41s

Provide feedback