[clang] [Clang][AMDGPU] Restore the non-RDC compilation pipeline (PR #200847)

[Yaxun Liu via cfe-commits]

yxsamliu wrote:

I opened #201135 as an alternative that keeps the non-LTO pipeline inside clang-linker-wrapper. The idea is the same goal (non-RDC shouldn't pay for LTO), but doing it in the wrapper means the per-arch device compiles still run in parallel, so `--offload-jobs` keeps working. Might be worth comparing the two approaches.


https://github.com/llvm/llvm-project/pull/200847