[llvm] [SystemZ] Don't lower float/double ATOMIC_[LOAD|STORE] to [LOAD|STORE] (PR #75879)

[Jonas Paulsson via llvm-commits]

JonPsson1 wrote:

As @arsenm suggested *getting rid* of ATOMIC_LOAD, while I thought the general idea was to clean up LOAD/STORE nodes handling by freeing them of possibly being atomic, I wonder if there is a consensus about this to follow...?

The reason I am asking is that it doesn't seem quite trivial after all to make sure all the reg/mem folding work both for LOADs and ATOMIC_LOADs in the SystemZ backend. 

https://github.com/llvm/llvm-project/pull/75879