Support for modified Semi-implicit euler solver

Hello! I'm trying to include diffrax to my wave simulation library as @patrick-kidger kindly suggested a while ago (and by the way, thanks a lot for this and lineax!) as I really would like to use the advanced checkpointing features 😄

However, my library also features a pseudo-spectral solver that requires a slight modification of the semi-implicit euler solver, where the discrete update equations also feature an extra $\alpha$ factor, that is:

$$
\begin{cases}
u_{n+1} = \alpha(\alpha u_n + \Delta t f(v_n)) \
v_{n+1} = \alpha(\alpha v_n + \Delta t g(u_{n+1}))
\end{cases}
$$

For the interested reader, this kind of integrator appears in equation 2.27 of this manual, and it uses the $\alpha$ term to correct the time-stepping error when absorption is added.

I don't think there's a way to do this with the current SemiImplicitEuler solver, so I wrote the modified version here: https://github.com/astanziola/diffrax/blob/semi-implicit-corrected/diffrax/solver/semi_implicit_euler_corrected.py

I'm very happy to make a PR, but before writing proper tests etc. I wanted to check first if this is really something that can't be done with the current solvers, and most importantly if it is of interest to add such a solver to the library.

Thanks!

Start by comparing the proposed diffrax/solver/semi_implicit_euler_corrected.py branch file with the existing SemiImplicitEuler solver and its tests. Confirm whether the extra alpha factor requires a distinct public solver and identify the relevant solver test entry points. Done means maintainers agree on the API and the solver has proper tests for the stated update equations.