there was an implicit assumption that the file exists before.

List of checkpoint files is now generated explicitly (rather than
listing files that match "checkpoint_*").
This fixes problems when encountering non-checkpoint files that do
contain the "checkpoint" word in their name.

initial condition must be generated even for unlisted solvers.
This commit also changes the name of a variable for style reasons

I think this line got lost in a merge somewhere... Not sure.

we're using clusters that require different slurm configs, hence the
need for fine grained control.

use less iterations, and only run one job.

this test puts 9 million particles on process 0, 1 million on process 2,
and 0 on the other processes (initially).
It's meant to test whether everything is working when some processes
don't have any particles.

There are now three targets for documentation:
`doc_doxygen` --- this runs doxygen on the CPP source code
`doc_html`    --- this runs sphinx on the python code, and it uses the
                  available doxygen xml output to generate
                  sphinx_html/...
`doc_latex`   --- this runs sphinx on the python code, and it uses the
                  available doxygen xml output to generate
                  sphinx_latex/...
I can't get it to a point where I can run just "make doc_latex", I need
separate runs of pdflatex, but I guess that's negligible.

Latex wasn't working because the README contains a link to an image that
cannot be converted. So I asked that the image only appears in html.
There were also some strange problems with the markdown math syntax, so
I switched to full LaTeX which seems to be working.