OpenMP target flags depend on CMake version

Nobody has claimed this yet.

In RAJA, we required CMake 3.14.5 as the min cmake version (default on LC systems). Then, we specified OpenMP compiler flags for OpenMP target offload builds using, for example for clang:

`-DOpenMP_CXX_FLAGS="-fopenmp;-fopenmp-targets=nvptx64-nvidia-cuda"`

In the last release, we bumped the min cmake version requirement to 3.20. Now, to build and link with OpenMP target offload, we need to use:

`-DBLT_OPENMP_COMPILE_FLAGS="-fopenmp;-fopenmp-targets=nvptx64-nvidia-cuda"
-DBLT_OPENMP_LINK_FLAGS="-fopenmp;-fopenmp-targets=nvptx64-nvidia-cuda"`

That is, we need to over-ride BLT logic for setting OpenMP complier and linker flags and force the flag settings to get the code to build and link. Fortunately, BLT gave us a way to do that.

Anyway, is this expected? Is there anything that can be done in BLT to simplify this cmake version dependency? Or, is there a better way we can handle this in our code?

The issue names BLT's OpenMP flag logic and the CMake 3.14.5 versus 3.20 behavior, but no files or tests. Start by locating the BLT logic that sets OpenMP compile and link flags, then compare the two CMake versions and the documented override variables. Done means establishing whether the dependency is expected and identifying a simpler supported configuration.

Written by the indexing model from the issue text.