Adaptive stochastic functionality

[maedoc]

There are a few related functionalities to adaptive stochastic schemes, which can be done with the Heun scheme we're accustomed to,

dx1 = f(x)
x1 = x + dt*dx1  # Euler predictor
dx2 = f(x1)
x2 = x + dt/2*(dx1 + dx2)  # Heun corrector

• error estimate, e.g. sum(square(x1 - x2))
• reject steps with error larger than some tolerance err > tol
• update step size dt *= tol / err and try again
• using Brownian bridge to "interpolate" the stochastic process between known sample values
  • also useful for looking at convergence of scheme w/ dt

Covered in detail by Rackauckas & Nie and their followup paper on stability optimized methods. The older Lamba technique is a useful warm up.

We might additionally want to support different dts per state variables per node, due to explicit time scale separations. It is also likely we should take into account different noise models per state variable, since applying the same noise to every state variable is not tenable for complex models.