Sparse (or kpm) implementation to avoid memory bottleneck

For large systems we suspect that memory will be a a significant bottleneck. This issue is a place to collect thoughts and ideas of how we can avoid this using a sparse mean field implementation.