Do we want the JAXopt- and Optimistix-based solvers to use the same tolerance and max number of steps if not specified?
Currently JAXopt solvers have their own defaults set in their __init__ in JAXopt.
Optimistix-based solvers use these defaults:
DEFAULT_ATOL = 1e-8
DEFAULT_RTOL = 0.0
DEFAULT_MAX_STEPS = 100_000
The potential problem with using the same for the two is that there is a difference in the convergence criteria between libraries. Optimistix uses what they call Cauchy criterion that uses both atol and rtol (which is disabled by default in NeMoS). JAXopt has a criterion defined per solver and uses a single tol.