Promote integral inputs to floating for `torch.logsumexp` (#63393)
Summary:
Fixed https://github.com/pytorch/pytorch/issues/56132, Integral inputs of `torch.logsumexp` would be promoted to the floating point type.
Pull Request resolved: https://github.com/pytorch/pytorch/pull/63393
Reviewed By: ezyang
Differential Revision: D30512180
Pulled By: mruberry
fbshipit-source-id: fbde3605c15b930411d0d1eb3a132b0088187097
