Single/modular precision fix ups

[sbryngelson]

Some fix-ups relevant to PR #632 by @arciyer123

• These cases are skipped in testing. I'm not sure why.
  • 1D -> 2 MPI Ranks, 2D -> 2 MPI Ranks, 3D -> 2 MPI Ranks
• Add benchmark check that ensures we see the expected speedup of single over double precision (and throw a non-zero stop code if we don't)

These are the two most pressing, I think. Everything else I can think of is more of a feature than something that needs to be fixed.

Update:

• I also see that the single precision tests don't test the examples. Not sure how that happened. It might have happened via a merged PR or perhaps always like that. Either way, we should look into fixing it.