[llvm] [LoongArch] Set isAsCheapAsAMove on LU{12,32,52}I/ADDI.D and {X}ORI instructions (PR #94733)

[via llvm-commits]

heiher wrote:

> Seems lu32i.d is 4x slower than move (or).

Real tests do not show `lu32i.d` to be slower than `move (or)` when they have the same data dependencies. However, I would still remove the `lu{12,32,52}i` instructions, as they add more instructions to the loop.

https://github.com/llvm/llvm-project/pull/94733