[llvm] [AMDGPU] Model waterfall loop EXEC update as a terminator (PR #219519)

[Arseniy Obolenskiy via llvm-commits]

aobolensk wrote:

Reworked the solution. Now we emit a plain andn2 plus a terminator EXEC write, and fuse them back after regalloc when safe

https://github.com/llvm/llvm-project/pull/219519