Generalize ODE solver

The ODE solver currently has two problems:

• it hard-codes the dynamics function
• it hard-codes the dimension of the state vector

This makes it tricky to experiment with things like, for example, adding an extra state to track angular distance travelled, or adding an extra parameter to the dynamics function.

I think that the rough design approach here would be to make a structure that holds the ODE solver cache data, which would allow us to avoid allocating new memory on each solver step. Then another utility function or class to actually use that ODE step solver to generate a complete simulation.

Rather than hard-coding the dynamics function, use rust's built-in features for high-performance lambda functions and a proper trait interface.

No files or tests are named, so first locate the current ODE solver and inspect how its dynamics function, state dimension, and per-step allocations are handled. Read the surrounding fractal simulation entry points before choosing the cache and simulation interfaces. Done means the solver supports varying state dimensions and dynamics callbacks through a Rust trait or closure interface without allocating on each step.