[llvm] [X86] Fold X * 1 + Z --> X + Z for VPMADD52L (PR #158516)

[Simon Pilgrim via llvm-commits]

https://github.com/RKSimon commented:

I'd be tempted to handle all this in combineVPMADD52LH where it should be cheaper:
 - Canonicalize constant multiplicands to the RHS
 - Handle `lo(X * 1) + Z --> lo(X) + Z iff X == lo(X)` pattern - nothing about it relies on SimplifyDemandedBits

https://github.com/llvm/llvm-project/pull/158516