statefulness of the model makes creating two solves from the same model more difficult

[yunlongxu-numagic]

See: https://github.com/numagic/lumos/blob/warm_start_with_model_solver/examples/laptime_simulation_example.py

One typical scenario is that we want to do an OCP on a model, and before the solve, we use ModelSolver to do some algebraic solves to warm start the problem.