[clang] [compiler-rt] [PGO][HIP] HSA-introspection device profile drain + GPU PGO tests (PR #203056)

[Larry Meadows via cfe-commits]

lfmeadow wrote:

@ronlieb @jhuber6 this adds a supplemental HSA-introspection
    device-profile drain so device-linked HIP programs with no host shadow (e.g.
    RCCL) get their device counters collected, plus the host-side
    clang_rt.profile_rocm linking fix for object-only links and a GPU-executed test
    suite. It's been validated end-to-end against a real RCCL build.


https://github.com/llvm/llvm-project/pull/203056