Allow diffusion to be specified as a constant

[mattlevine22]

Currently ContinuousTimeStateEvolution requires diffusion to be lambda x, u, t: Callable.

1. This makes writing models with constant diffusion (kinda the default use case) unnecessarily cumbersome
2. Makes it harder for us to exploit things like scalar / diagonal diffusion coefficients.

Could do some nice auto-broadcasting for hierarchical cases as well here.