Second-order in time

[rckirby]

It should be possible to extend Irksome to do second-order in time problems via Runge-Kutta-Nystrom without forcing a rewrite as a first-order system (which uses more memory to store solutions and requires a larger stiffness matrix per-time-step). Things that have to happen:

• Extend TimeDerivative to allow higher-order derivatives
• Allow expanded tableaux for RKN
• Figure out how to modify the UFL substitutions to DTRT, making sure we can do first and second derivatives in time (e.g. telegraph equation)
• Engineer the top-level interface. Plausibly we want TimeStepper to uniformly accept second-order equations, just making sure that we have a ButcherTableau suitable for RKN on input.