testing PolonskyKeerRey in MPI

It seems to work, although there's a remainder error in the comparison between
the sequential and the parallel solves. It is possible that the change in the
order of reduction operations changes the solution. Needs further investigation.