[llvm] [AMDGPU] Lower `llvm.amdgcn.queue.ptr` instrinsic to using implicit kernel argument if feasible (PR #103490)

[Matt Arsenault via llvm-commits]

arsenm wrote:

> > There is already a representation for load from implicit argument pointer, and we don't need 2 ways of doing that.
> 
> Which is the representation for loading queue ptr from implicit kernel argument?

Call the implicitarg ptr intrinsic and load from that offset 

https://github.com/llvm/llvm-project/pull/103490