Check binding of threads and processes for GROMACS

[casparvl]

Currently, we don't set any binding options explicitly for the GROMACS test. This may or may not result in 'sensible' binding. We should check how we can make sure that it binds in a reasonable way. Also, we should check that it launches correct amounts of tasks on hyperthreading nodes (1 per physical core, not 1 per thread).

For binding, we'd want processes to be bound to physical cores for pure MPI (i.e. CPU runs of GROMACS). For GPU runs (essentially hybrid OpenMP+MPI) we should bind at least the tasks to 1-xth of the nodes CPU cores (if there are x GPUs in the node). Even better would be to also add thread binding to this.