Partitioned RK methods

[rckirby]

There is a literature on interesting partitioned methods for problems of the form p' = f(p, q); q' = g(p, q) where one uses different but compatible tableaux to discretize the two equations. It would be interesting to support these -- I think the UFL manipulation would be quite similar to what we've got, but there might be interesting solver questions, especially when one computes one stage at a time.