[llvm] [AMDGPU][NFC] Added Pre-commit tests for PR#137137 (PR #137150)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> Here, the main differentiation should be vectors of fp16 (having different element numbers), in order to see how packing is handled in different scenarios.

All of the base intrinsic tests should cover all the interesting number of vector elements. This isn't covering any different source type packing situations, these are all just direct intrinsic uses. 

https://github.com/llvm/llvm-project/pull/137150