Implements the change proposed in #193 (closed).
Triple benefit:
- Stats are computed at the correct point in time and are hence more accurate.
- The drift does not compute stats anymore so this might help with the bandwidth.
- The cell structure becomes 96 bytes lighter which will also help with the bandwidth.
The stats are now computed when requested by the used using a threadpoolized loop over the particles. This happens rarely and is hence not a burden on the whole system any more. The interesting stuff is in src/statistics.[ch].
This is built on top of mark_task_in_drift2 so should merge smoothly with master very soon.