Revisit MPI after / with #149

[jkozdon]

Since a lot of this will be rewritten with #149, I am going to merge this so that @nicoberlie can build on this for #119.

There are a few hiccups with this code, but I think that this will get sorted out with #149 and subsequent PRs which have more rigorous testing.

All of the problems are related to MPI (either MPI + windows or MPI + Ubuntu on the example). My hunch is that there is a garbage collection issue going on, but I think (hope?) that better testing with MPI and library the reworking will sort these problems out.

Going to open an issue to track this so that we remember to revisit it.

Originally posted by @jkozdon in #120 (comment)