Return contiguous tensor from softmax decomposition (#85788)
Fixes https://github.com/pytorch/torchdynamo/issues/1135
Softmax decomp's output stride does not match with aten softmax output stride. Not sure if its desirable. Opening a PR for now.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/85788
Approved by: https://github.com/ngimel, https://github.com/ezyang
