Implement StateSpace class hierarchy

[dglmoore]

The StateSpace class is very generally. For example, an instance can represent a state space in which each dimension has a different base (useful for models with nodes of different bases). Generality is an advantage in that is means we can use it in a lot of situations, but it sometimes comes at a performance cost. General functions cannot always make assumptions that would otherwise improve performance. For example, the StateSpace.encode method doesn't make use of any bit-level operations because there's no guarantee that the node-states will be Boolean.

Is it worth implementing a class hierarchy of state spaces that allow more efficient implementations of basic algorithms?

Proposed API

The hierarchy starts at the top-level with a maximally general state space.

class StateSpace(object):
    """Each dimension can have a different base, not necessarily Boolean"""
    pass

Lower and lower levels have more specific constraints on them:

class UniformStateSpace(StateSpace):
    """Each dimension has the same base, not necessarily Boolean"""
    pass

class BooleanStateSpace(UniformStateSpace):
    """Each dimension is Boolean"""
    pass