Implementation of allowed states in the game abstract class

[shev72]

Currently all state transitions are allowed, it would be nice in the base class if we could override a method which allows us to return the set of allowed next states. Currently this must be learned by the algorithm by using the score. I have forked the code and may be able to propose an implementation, I think it is not too hard, I need to study the core training class to see how to implement this.