[clang] [llvm] AMDGPU: Implement tensor load and store instructions for gfx1250 (PR #146636)

[Sameer Sahasrabuddhe via cfe-commits]

ssahasra wrote:

To wit, those _non-async_ variants were added in #146636, without much of a review or a discussion of what async means. It's that actual practice of silently lifting things from NPI to upstream which needs to be discontinued.

https://github.com/llvm/llvm-project/pull/146636