[CUDA] Enable tests in AtomicRef for CUDA after adding support for generic address space in DPC++

[ghost]

Currently the tests in AtomicRef for the generic address space are separated in different *_generic.cpp files just to let us disable them on CUDA. After support for the generic address space is added to the CUDA backend, the tests should be enabled and can be merged into the .cpp files for the tests with the global address space.

See also: #5215

[tomflinda]

Could you provide the time when support for the generic address space is added to the CUDA backend?

[t4c1]

I would prefer to wait until existing atomic related PRs are merged (especially intel/llvm-test-suite#534 and intel/llvm-test-suite#648), so we have good tests when we are making these changes. Other than that, the changes should be relatively simple.

[rodburns]

Fixed by intel/llvm-test-suite#1446