Implement memory-efficient way to handle observables during run

[RandomDefaultUser]

This is not critical at the moment, because we calculate stuff like the energy and pressure, but if we would like to also calculate e.g. RDFs or the stress tensor (and save it in a meaningful way) we cannot hold them in a numpy array during the run and then save them at the very end. This is incredibly memory-inefficient. What we would need is an interface that allows us to append data to some binary file format, e.g. with zarr. That way we could continuously update some data files.