Turns out that having blocking MPI calls in the dumper code is bad as that can be blocked by other MPI calls in other threads, so we need to avoid any MPI calls here and create a file per rank instead

Make the new task synchronize the particles hit by feedback and force them to recompute a time-step. Propagate the time-step upwards.

Silly

Interesting to see what tasks complete first and last

Polls for a file .dump and when found dumps the active tasks and memory (if configured)

Version that correctly flags the BHs to swallow and removes them before dying in the time-step calculation function.

Split the star formation task into two parts: The star formation bit and the resorting bit. Move the sf_count communication to just depend on the first one in order to speed things up.

Added communication functions to propagate the new number of star particles after a SF event has occured.

Added basic support for black hole particles. Particles can be read in, attached to gpart, integrated forward in time, drifted and written to snapshots.

Added a force-communication with its dependency on the drift to make sure the data is correctly drifted before being received in the star loop.