No longer bit-for-bit deterministic with >1 MPI rank

[bprather]

The current regression tests force a single MPI rank when testing reinitialization and restarting. This is because with more than 1 rank, the results are not bit-for-bit identical -- something about the ordering of receiving MPI buffers in Parthenon or KHARMA can affect round-off. This occurs only with Face-CT transport, which also enables flux corrections in Parthenon, so this seems likely to be the culprit.