Add a configured header instead of relying on cmake defines

[adayton1]

In umpire, after importing camp using find_package, this bit of cmake is required to get a CUDA build of umpire working:

blt_add_target_definitions(
      TO camp
      SCOPE INTERFACE
      TARGET_DEFINITIONS CAMP_HAVE_CUDA)

Could we add a configurable header that handles this issue and precludes inconsistency across projects using camp?

[davidbeckingsale]

Just for a bit more context, the CAMP_HAVE_CUDA macro is automatically defined when using nvcc:

camp/include/camp/defines.hpp

Line 67 in eb555ee

#define CAMP_DEVICE __device__

However, in Umpire we are only linking to the CUDA runtime since we don't actually have device code.