[llvm] [AMDGPU] Add an asm directive to track code_object_version (PR #76267)

[Emma Pilkington via llvm-commits]

epilk wrote:

> > @slinder1 @arsenm Ah, so code_object_version is only meaningful on HSA OSes? I originally chose 'amdgcn' since I thought it was also meaningful on PAL/Mesa. The update renames the directive to `amdhsa_code_object_version` and only emits/parses it on HSA triples.
> > I noticed that clang emits the IR module flag, `amdgpu_code_object_verison` on PAL/mesa. Should that be updated to just emit on HSA? Thanks!
> 
> It's not exactly well defined enough to make a definitive statement. The hope was mesa would follow HSA code objects, but that maintenance work never happened. Less sure about PAL

Hmm, okay. So would you be okay with the the new approach of using the `amdhsa` spelling and requiring a HSA triple? I suppose we could always add support for PAL/mesa if we need that in the future.

https://github.com/llvm/llvm-project/pull/76267