[llvm] [NFC][AMDGPU] Move `flat_atomic.ll` to `llvm/test/CodeGen/AMDGPU/` (PR #141126)

[Shilei Tian via llvm-commits]

shiltian wrote:

yes it looks like running end-to-end tests as it uses `llc`'s default optimization level.

https://github.com/llvm/llvm-project/pull/141126