[llvm] [AMDGPU][GlobalISel] Don't legalize scalar s16 merge/unmerge into wider scalars with true16 (PR #207999)

[Petar Avramovic via llvm-commits]

petar-avramovic wrote:

Need ll test. Why not select into 16bit sub-register index copy?
I assume this will end up with 32bit then 16bit sub-register index copy.

https://github.com/llvm/llvm-project/pull/207999