[SR] Fuse clamp/nan_to_num
Pull Request resolved: https://github.com/pytorch/pytorch/pull/77094
Fuse `clamp` and `nan_to_num` in an NNC kernel. This leads to a big speed up on many models. We can avoid comparisons since clamp potentially gets rid of all of the `inf`s in the input tensor.
Differential Revision: [D36220967](https://our.internmc.facebook.com/intern/diff/D36220967/)
**NOTE FOR REVIEWERS**: This PR has internal Facebook specific changes or comments, please review them on [Phabricator](https://our.internmc.facebook.com/intern/diff/D36220967/)!
Approved by: https://github.com/navahgar
