Control optimization

[jcrozum]

The control algorithm can be improved slightly.

1. A path can be ended early if all descendants of a node that have attractors correspond to subspaces of the target subspace.
2. We may as well have the user pass in a succession diagram object only because it stores the rules inside. Right now, we have two copies of the rules being used.

Point 1 requires knowing what attractors are reachable from each SD node, so it's only viable if we pass in an SD that already has all the attractors identified.

[kyuhyongpark]

I wonder if point 1 requires the SD to be fully expanded (bfs or dfs) or not. @daemontus would the attractor seed expansion method ensure that every SD node has at least a single path to each of the attractors it can reach? I think this is not the case for at least the minimal trapspace guided expansion (ignoring motif avoidant attractors), since it stops expanding to minimal trapspaces that already has a single path.

[daemontus]

I believe that point (1) does not require full expansion, but it depends on the NFVS results: If the NFVS method confirms that all attractor candidates are in the target subspace, we can end the path. Hence we don't even have to compute all attractors in the subspaces, we just need to rule out or confirm any candidates that are outside of the target subspace?