[AMD] Fix broken build from nested transformer utils (#110245)
Summary: D49374910 breaks internal amd build because we didn't hipify the header file in nested/cuda. Maybe it's just easier to move it outside.
Reviewed By: nmacchioni
Differential Revision: D49743234
Pull Request resolved: https://github.com/pytorch/pytorch/pull/110245
Approved by: https://github.com/drisspg
