Added an (expensive) debugging check to verify that all cells have been drifted to the current point in time before a rebuild or a repartition.

first try to merge in the scheduler_active stuff from the scheduler_skip branch. currently broken, apparently engine_collect_timestep isn't doing anything.

All active cells must be drifted so do not skip them if rebuilding.

If rebuilding no need to unskip tasks, that will be after the rebuild

Make sure we visit all related sends of the tasks, not just of the cell

Moved the un-skipping of tasks to a separate function. Abort drifting when forcerbuild is on. Access forcerebuild atomically.

Promoted the external gravity tasks to a self sub-type such that the super pointer can be set correctly.