Merge Trilinos support

[Jack Hale]

Would be good to just add Trilinos to all the current builds, rather than having separate builds for PETSc/Trilinos users. @chris_richardson is this reasonably stable and ready?

[Prof Garth Wells]

If we're happy with the increase in the image size, I'd suggest adding Trilinos to the lowest possible level image.

[Chris Richardson]

I'm just going to try it as a separate image on dockerhub first...

[Jack Hale]

Oh dear, think we've reached the 2GB RAM limit on Dockerhub's build instances.

https://hub.docker.com/r/fenicsproject/dev-tpetra/builds/b8xxpde9gghs5stvb3yfdca/

Must be all those tasty templated classes in Tpetra.

[Chris Richardson]

Yes, I just changed make -j 4 to make -j 1 - maybe it will help.

[Prof Garth Wells]

Probably best to compiler with just one thread on free, hosted systems.

[Chris Richardson]

Well, that did't work either...

[Prof Garth Wells]

Wow, does it really need more than 2GB RAM to compile!

[Jack Hale]

This will have to remain a non-automated build for the time being. I will build it on my laptop and push it to dockerhub manually. It would be nice if dockerhub offered a 'premium build' service with bigger amazon build slaves.