[llvm] [clang] [AMDGPU][GFX12] Add new v_permlane16 variants (PR #75475)

[Matt Arsenault via cfe-commits]

arsenm wrote:

> Looks like the new test might be failing on macOS/arm64: http://45.33.8.238/macm1/75171/step_11.txt

> Do you want to switch the test to use `-mtriple`, or do you want to revert and investigate what's up with the assert?

Really it needs #75469 for the triple to not default to a nonsensical combination 

https://github.com/llvm/llvm-project/pull/75475