Potential changes for a breaking release

[isaacsas]

It might make sense to layout some potential updates here that would make the workflow nicer and more consistent with other SciML libraries, but be breaking. JumpProcesses has a lot of historical cruft at this point which could potentially be cleaned up with a redesign. Some examples:

• Specifying VRJ aggregators.
  • We essentially have a FRM currently, what I proposed the other day is a DM. There is also Anderson’s MFRM method and others. Plus the mix of integrating via the ODE integrator vs. quadrature. But selecting solvers here would require an interface.
  • To eventually support hybrid models it might make more sense to just have some type of structure that stores / encapsulates multiple solver selections (for constant jumps, variable jumps, ODEs, SDEs, etc), and knows how to put them together into a single solver.
  • For pure jump systems that have VRJs it seems like we might benefit from a version of SSAStepper or such and avoid ODE integrators (i.e. just have a time-stepper with root finding).
• Not passing or building aggregators at problem construction (removing mutable state in JumpProblems). i.e. defering aggregator selection until calls to solve.
• Storing rngs in integrators instead of in aggregators/JumpProblems (and providing users an API to access them that we internally use too).
• Not having solve reseed when no seed is passed, but leaving such behavior to higher level callers.
• reinit support to avoid overhead of reconstructing solvers if making changes as mentioned above.
• Have aggregators store their kwargs, i.e. instead of passing dependency graphs as arguments to JumpProblem pass them as arguments to constructors of aggregators, i.e. NRM(; dep_graph), RSSA(; variables_to_jump_graph), etc.
• Nailing down the bounded VRJ interface (should we have one function that returns all bounds and a time window, a function that just does upper bounds and time window, leave alone as separate functions as we do now)?

edit: This is just a draft, but I'll try to iterate on over time.

[ChrisRackauckas]

All of these look like reasonable changes. I presume we might want a few changes to regular jumps / tau leaping as well, but I don't have any concrete suggestions right now.

For pure jump systems that have VRJs it seems like we might benefit from a version of SSAStepper or such and avoid ODE integrators (i.e. just have a time-stepper with root finding).

So a case where they are all time-dependent but not defined by an ODE? That would be fine but I haven't seen an example of that yet.

[isaacsas]

So a case where they are all time-dependent but not defined by an ODE? That would be fine but I haven't seen an example of that yet.

Modeling the changing volume of a cell is a case where ODEs aren't really needed.