[tensorexpr] support for fusing autocasting ops (#72478)
Summary:
Pull Request resolved: https://github.com/pytorch/pytorch/pull/72478
aten::_autocast_to_reduced_precision and `aten::_autocast_to_full_precision are essentially just aten::to operations, so they can be fused the same way aten::to is fused.
Test Plan: Imported from OSS
Reviewed By: bdhirsh
Differential Revision: D34057522
Pulled By: davidberard98
fbshipit-source-id: f3b53641415702a4ac56460587801b9c76d81b3c
(cherry picked from commit 838ce5542ebd1f95be1ab57ed496f16db4fb7d68)
