[clang] [CIR] Wire AMDGPU into the call-convention lowering pass (PR #220197)

[via cfe-commits]

skc7 wrote:

> I'm planning a major refactoring of the calling convention lowering pass to use inheritance to separate the target independent parts from the target-specific parts, but I've been waiting for the x86 implementation to stabilize. I expected there to be more x86-specific logic that you'd need to exclude. Did you intend this as just a first step?

Yes, this is first step that wires the AMDGPU classifier into the existing pass.. Happy to rebase onto your refactoring.

https://github.com/llvm/llvm-project/pull/220197