Change !545 breaks mdrun without MPI or THREAD_MPI, and all reference builds
The change !545 (merged) that should "Improve naming and docs" introduces changes in the logic of communicators that causes all Gromacs builds where both MPI and THREAD_MPI are disabled to always fail with a critical assert at the start of mdrun.
The relevant change seems to be around line 852 in runner.cpp where communicator is now checked to be different from MPI_COMM_NULL, rather than identical to MPI_COMM_WORLD.
Since the reason for changing the logic isn't documented in the comment, I presume this can be reverted tomorrow unless we hear back from @eirrgang until then.
This is currently making it impossible to update our reference values in the regression tests (since those need a reference build type, where all MPI options are disabled), which in turn is holding up %2021.4, so it's pretty urgent.