Add tests for aten::record_stream

[Kanya-Mo]

Currently there's no available UT in pytorch to test record_stream. These two tests are adapted from corresponding cuda tests. The only difference is I use an actual expensive kernel in place of torch.cuda._sleep to create delay in one stream. The add kernel here would create sufficient delay based on max memory bandwidth among current supported gpus.

[xytintel]

@Kanya-Mo Pls move src/ATen/native/xpu/sycl/RecordStream.cpp to src/ATen/native/xpu/RecordStream.cpp

[Kanya-Mo]

@Kanya-Mo Pls move src/ATen/native/xpu/sycl/RecordStream.cpp to src/ATen/native/xpu/RecordStream.cpp

Moved