IMEX

[rckirby]

We have a trial IMEX-type method for Buvoli's Radau-based schemes. A more general implementation that lets us do additively-split methods for u_t + F1(u) + F2(u) = 0 where one term is handled implicitly and the other explicitly. This requires supporting a pair of tableaux and doing some UFL manipulations. Most classical IMEX methods are only diagonally implicit, so there should be a way to adapt/extend our DIRK stepper for these.

[rckirby]

This was done in #106