In-State Variation of Solver Characteristics

As I continue to design a Leap integrator that provides characteristics as close to CVODE as possible, an issue I continually encounter is the fact that the use case I'm targeting involves performing the implicit integration - which in the case of CVODE is fairly complex in terms of control flow - on a point-by-point basis. Given that we wouldn't want to call any Leap integrator this way, a natural difficulty that arises is that having point-local data in an adaptive implicit solver means that integrator characteristics like order and timestep become point-local. This is not something I know how to describe using Leap and Dagrt, nor is it something I think we're willing to part with in terms of capability for our use case.

Assuming a state that's comprised of point-local chunks (arrays?) of data, one way to go would be to change the UserType - or add a different one - that can better accommodate this use case. Open to any feedback/suggestion on this, and Andreas if I didn't explain the issue clearly feel free to provide more context.