Running on HPC

[robertodr]

The SLURM script shows the options that work for Beskow, but these might not be universally applicable. We should probably add:

• short discussion of differences between srun, mpirun, and mpiexec as launchers for MPI jobs.
• why one should prefer srun (if it's correctly set up by the sysadmins on the cluster)
• explain how one might use --map-by and --bind-to and how they can affect performance.
• add a strong suggestion to always benchmark the launcher options for a given cluster. This could evolve into a short performance tuning guide to include in the official documentation.