Add RATTLE solver

[krzysztofrusek]

This PR adds RATTLE solver for constrained Hamiltonian dynamics. It's very similar to #303 with the addition of implicit constrain.
I use this solver for HMC on implicitly defined manifolds in jax_chmc and it may be useful to have it in diffrax.

This is the first solver with constraints so I am not sure if the abstraction is correct.
Since the constrain function is used in the solver step, maybe it's better to introduce another type of Term?

[patrick-kidger]

Thank you for the contribution!

The introduction of a constraint here is pretty interesting. As it happens, I'm actually aiming to have Diffrax support semi-explicit DAEs at the diffeqsolve level -- for all kinds of solvers. (I'm hoping to get that in in a couple of releases, maybe around the new year.) Given that, I'm thinking it might make most sense to hold off on merging this until I've got a draft of that going -- I can tag you on that to get your thoughts once it's done.

WDYT?

[krzysztofrusek]

Ok, lets wait for you to finish and I would love to give it a try - epscially for symplectic integrators.