DM and DMDA interface

[jkozdon]

@nicoberlie and @boriskaus: This is the DM / DMDA interface I started working on a while ago. Some similar functionality to what you two have done, but not DMStag stuff.

Not saying we need to use what I've done, just wanted to throw it up so you could see some of the choices I made.

(This was done before I really started to dig into the guts of everything, so I might make some different design choices now.)

To wrap still:

• DMCreateGlobalVector
• DMCreateLocalVector
• DMLocalToGlobal
• DMGlobalToLocal
• DMCreateMatrix
• DMGetCoordinateDM
• DMGetCoordinatesLocal

[jkozdon]

@nicoberlie, @boriskaus, @psanan

I think I've mostly finished up the broad reworking of this PR to start to add the DM and DMDA interface. Would be interested in your thoughts on how it feels.

I struggled a bit with whether to match the PETSc names or make things a bit more "julian". I went with the latter since that seems to be what was done before and it just reads a bit nicer to me.

(I didn't touch yet any of the other parts of the code, such as Mat, Vec, KSP, etc.)

Feel free to 🚲 🏠 (bikeshed).

A good place to jump in might be examples/ksp/ex50.jl and/or test/dmda.jl

[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.