note: the sphysics sim does a recount every 100 timesteps on which particles are there and not - this means number of particles is changing
thought i'd be able to implement that today but alas, too much complix going on
might do fresh start on that or just revert
BUT: so far, single file with no unlimited dims is MUCH faster, and 15th of size of previous output file (with 2/7 of output, so still at least halved file size?)
other stuff for tomorrow/next time
next on to-do list: netcdf-python
hack an input file manually to see how paraview copes with changing number of unstructured grid points!