[llvm] [Offload] Implement the remaining initial Offload API (PR #122106)

[Joseph Huber via llvm-commits]

jhuber6 wrote:

What is the plan for testing things on the GPU? I'm thinking it would be best to store some C/C++ code that tests the desired functionality and build them separately to a specific name, then mark it as a dependency on the unit test target and just open it as file. I.e.

```
#include <gpuintrin.h>

extern "C" __gpu_kernel void kernel() { some_test(); }
```

Another thing we need to fix in the `libomptarget` runtime is that the images are assumed to live throughout the entire execution, we really need to copy it internally and maintain its lifetime ourself now that it's going to come from external users instead of a constant pointer in an ELF.

https://github.com/llvm/llvm-project/pull/122106