[llvm] [LoongArch] lower vector shuffle to shift if possible (PR #132866)

[via llvm-commits]

heiher wrote:

> it's not work for LASX now

Will this optimization be added for `LASX` in a future patch, or is there a chance it could be included in the current one?

https://github.com/llvm/llvm-project/pull/132866