Add 1-layer gradient accumulation test to check aliasing #9365

Job	Run time
get-torch-commit	4s
Build XLA CUDA plugin / build	5m 8s
Build PyTorch/XLA / build	22m 44s
Build PyTorch with CUDA / build	24m 13s
TPU tests	0s
CPU tests / test (benchmark_tests)	8m 1s
GPU tests / test (benchmark_tests)	8m 57s
Build docs / build-docs	1m 34s
CPU tests / test (python_tests, xla_op1)	8m 54s
GPU tests / test (python_tests, xla_op1)	8m 58s
CPU tests / test (python_tests, xla_op2)	4m 4s
GPU tests / test (python_tests, xla_op2)	7m 39s
CPU tests / test (python_tests, xla_op3)	5m 58s
GPU tests / test (python_tests, xla_op3)	8m 55s
CPU tests / test (python_tests, torch_mp_op)	8m 59s
GPU tests / test (python_tests, torch_mp_op)	8m 57s
CPU tests / test (cpp_tests, cpp_tests1)	4m 52s
GPU tests / test (cpp_tests, cpp_tests1)	8m 58s
CPU tests / test (cpp_tests, cpp_tests2)	5m 1s
GPU tests / test (cpp_tests, cpp_tests2)	9m 1s
GPU tests requiring torch CUDA / test (python_tests, linux.8xlarge.nvidia.gpu)	7m 33s
GPU tests requiring torch CUDA / test (triton_tests, linux.g5.4xlarge.nvidia.gpu)	5m 29s
	2h 53m 59s

Provide feedback