Backtracking during Restart

[ImkoMarijnissen]

In the case of a restart, it could be beneficial to completely recalculate from scratch rather than to notify of the (potentially) numerous events.

This would require:

• Determining when a restart loses "too much" information; we could use a heuristic based on how close the current state of the solver is to a solution
• Having an interface for propagating from scratch for every propagator
• Ensuring that the stored events are cleared properly

[maartenflippo]

Why do we not currently support this? We could say that when we are restarting, we don't notify on backtrack events. Then the synchronize will be called which should be implemented such that the propagator's state is correct wrt the current state.