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

[Joseph Huber via llvm-commits]

jhuber6 wrote:

Worked when I tested it. But in the future running the `ninja` target needs to run the test with a target like `ninja check-offload-unit` and should auto detect the GPUs on the system.

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