[llvm] [Offload] Add MPI Proxy Plugin (PR #114574)

[via llvm-commits]

================
@@ -6,6 +6,7 @@
 // UNSUPPORTED: amdgcn-amd-amdhsa
 // UNSUPPORTED: nvptx64-nvidia-cuda
 // UNSUPPORTED: nvptx64-nvidia-cuda-LTO
+// UNSUPPORTED: nvptx64-nvidia-cuda-mpi
----------------
jprotze wrote:

If a specific testsuite is not supported, I think this is the right notion. For most of the tests touched in this PR, I think they should just use `cuda`. Or do they really succeed for `nvptx64-nvidia-cuda-JIT-LTO` and not for the other cuda targets?

Also: UNSUPPORTED means that the tests are executed and must fail. Is this what we want to test here, or should they use `REQUIRES: cuda` and skip the test for this platform?

https://github.com/llvm/llvm-project/pull/114574