Add 1d support for ArborXWrappers

[masterleinad]

Related to #15538.

[Rombur]

We now require ArborX 1.3 which has multi-dimensional search support. It's probably cleaner and faster to use the new interface instead of copying everything to 3D data structures.

[masterleinad]

We now require ArborX 1.3 which has multi-dimensional search support. It's probably cleaner and faster to use the new interface instead of copying everything to 3D data structures.

That requires incompatible changes to the predicates, though, since they are not templated on the dimension but store 3-dim objects. I would rather do that in a follow-up.

[Rombur]

Using CTAD, adding a template on the dimension should be backward compatible because the constructor of the predicates takes a Point or a BoundingBox that is already templated on the dimension.

[masterleinad]

Using CTAD, adding a template on the dimension should be backward compatible because the constructor of the predicates takes a Point or a BoundingBox that is already templated on the dimension.

I had a look at how much it actually takes to use the ExperimentalHyperGeometry namespace. From a user perspective, i.e., in the examples everything works fine without changes due to CTAD. Deriving your own predicates or primitives from the ones provided is not backward-compatible, though, since you have to provide the additional template argument. What's more, is that Morton codes are not defined for dim=1 and the distributed BVH implementation is not templated on the type of the bounding volume hierarchy but uses the BVH class (using ArborX::Box) for both the upper and lower tree.
Thus, I think we need a couple more changes in ArborX before it really makes sense to switch to lower-dimensional data structures here.

[Rombur]

clang-format is not happy