Abstract Looping

[bprather]

Currently, loops in k-harm are implemented directly as Kokkos::parallel_for calls given MDRange objects. This specifies what should be looped over, but also how it should be done. Ideally these should be split: some abstraction should be given an MDRange object, and then decide how (or even whether) to call Kokkos.

This is probably easiest to do by coding a drop-in replacement for parallel_for, which then looks inside the MDRange and decides what to call inside. This will need to be templated on the MDRange Rank.