Fixes #488 (closed).
This possibly fixes an old bug that would seldomly lead to unsorted cells in the pair interactions. #488 (closed) was the first time I could catch it in an easily reproducible scenario.
Does the fix in the recursion of runner_do_sort() make sense to you as well? The idea is that we set the sort flag in the unskip based on the old value of dx_max_part so we should use that same value here and not the one obtained after the drift that just happened.