[llvm] [AMDGPU] Vectorize more 16 bit shuffles (PR #90648)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Alive2 complains about this patch:

This patch can't have broken this as it just changes the cost model, this must have just added test coverage that hits it 


https://github.com/llvm/llvm-project/pull/90648