Understand the overhead of LinkedCellsReferences

Understand the potential overhead of switching from the current container to reference-based containers.

Currently, with AoS, we see roughly the same performance between LinkedCells and LinkedCellsReferences (even without sorting). This is strange. Furthermore, even with sorting the difference in performance is small.

Investigate these effects by looking at cache misses vs runtime in the relevant parts of the simulation.

The conclusion of this issue is not some contribution to the code, but to understand fully the performance implications of reference-based containers, which is an important first step in the milestone Kokkos.