Refine storage of subsolver states

[kellertuer]

See #399 (comment),
we could store the evaluation mode of a closed form solution also in a small state to make the type of sub_state more unified.
The keyword should still accept just the evaluation type.

This sounds like an easy refactor task.

[mateuszbaran]

By the way, do you have an example of where a closed-form subsolver is used?

[kellertuer]

Frank-Wolfe often, Difference-of-Convex, maybe ALM sometimes?

In general it might also be nice to pass a “closed form solution” function to use some other packages solver as a subsolver.