Improve data life time management

[Sichao25]

PCMS need to clarify where ownership and copying of data are actually required.

A general question is, conceptually, PCMS field evaluation should not need to own the data itself. When a field is created, the underlying data and coordinates likely already exist elsewhere. However, I am not sure if it is practical to completely avoid copying since we need local array eventually for coupler. At least we should try to use the same data in each memory space

Another specific case can be improved is about CoordinateView usage. For example,

pcms/src/pcms/field/evaluator/mesh_fields.h

Lines 131 to 142 in 3325cb2

	auto coordinates = coords.GetCoordinates();
	Kokkos::View<Real* [2], DeviceMemorySpace> coords_search(
	"coords_search", coordinates.extent(0));
	Kokkos::parallel_for(
	"copy_coords",
	Kokkos::RangePolicy<DeviceMemorySpace::execution_space>(
	0, coordinates.extent(0)),
	KOKKOS_LAMBDA(const int i) {
	coords_search(i, 0) = coordinates(i, 0);
	coords_search(i, 1) = coordinates(i, 1);
	});
	auto results = search_(coords_search);

CoordinateView takes no ownership, is used in evaluation requests and evaluators, but point search requires a Kokkos view — forcing an internal copy of the coordinates inside the evaluator. This is not ideal and potentially avoidable pattern worth addressing.

[jacobmerson]

Yes, this is problematic. If we update the search to take a ArrayView that should resolve this issue correct? The search shouldn't store the coordinates. I think the search API just never got updated to use ArrayViews.

I will further note, that only recently do we have all these pieces working together, so some of these inconsistencies are more due to pieces being independently developed, than specific design choices.

I do appreciate that you are documenting these issues because many of them should be fairly straightforward to clean up and will be very beneficial for performance / readability.