Implement self-adjoint time discretization schemes

[agzimmerman]

Right now we're only using fully implicit backward difference formulas of first, second, and third order. Thankfully, the time discretization of any of our models is well encapsulated, so we can test mixed explicit-implicit methods such as Crank Nicholson and Runge-Kutta. @tomoleary said that Crank Nicholson and RK4 are needed for plugging our models into inverse problems because they are self-adjoint.