[clang] [llvm] [OffloadBundler] Rework the ctor of `OffloadTargetInfo` to support AMDGPU's generic target (PR #122629)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

> Isn't offload bundler a target agnostic tool?

Would it break anything if we just assumed it was always `amdgcn-amd-amdhsa` for the triple? It's supposed to be generic, but right now only HIP uses it AFAICT.

https://github.com/llvm/llvm-project/pull/122629