[llvm] [OpenMP] Add test to print interop identifiers (PR #161434)

[Jan André Reuter via llvm-commits]

Thyre wrote:

The strings for NVIDIA look good:

```console
$ clang --version
clang version 22.0.0git (https://github.com/adurang/llvm-project.git 6090c32a072bd97fc24709c91ffe6dde770e0577)
Target: aarch64-unknown-linux-gnu
Thread model: posix
InstalledDir: /p/project1/ccstpa/reuter1/Projects/Sources/LLVM/interop/llvm-project/_build/_install/bin
$ ./bin/clang test.c -fopenmp --offload-arch=sm_90 -frtlib-add-rpath -L$(pwd)/_install/lib/aarch64-unknown-linux-gnu -Wl,-rpath=$(pwd)/_install/lib/aarch64-unknown-linux-gnu
$ ./a.out
omp_get_interop_int returned cuda
omp_get_interop_str returned nvidia
omp_get_interop_str returned cuda
"PluginInterface" error: Failure to synchronize interop object 0x0000aaaaeccff8c0: "generic error code for features unsupported by the device/backend" sync_barrier not supported
omptarget fatal error 0: Interop sync barrier failed for 0xaaaaeccff8c0 object

Aborted (core dumped)
```

The error at the end should be expected if I understood the discussion in the other PRs correctly. Switching to `target` instead of `targetsync` still yields the same strings, but removes the error at the end.

---

Regarding multi-GPU with different archs, _maybe_ I can check this but I can't guarantee that I get this to work.

https://github.com/llvm/llvm-project/pull/161434