[llvm] [AMDGPU] Reimplement V_READFIRSTLANE_B32 as a normal VOP1 Pseudo. NFCI. (PR #81877)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> This was a weird case that didn't have the usual Pseudo/Real abstraction.

There was a reason for this that I don't really remember. I think the legal source operands changed slightly at some point 

https://github.com/llvm/llvm-project/pull/81877