Add ability to collapse Observable space from states

[sritchie]

If you have some state that returns a really rich observation, you may not want to key your value function on the entire thing. You might want to discretize the angles etc in the Cartpole example, for instance.

We should add the ability to... map the observable, maybe on the state itself? think it through.

[sritchie]

See the PoleCartState vs RoughPoleCartState, and check out StateConversion[PoleCartState, RoughPoleCartState]...

https://github.com/cb372/reinforcement-learning-in-scala/blob/master/src/main/scala/rl/polecart/core/PoleBalancingProblem.scala#L156

The example does that like this:
https://github.com/cb372/reinforcement-learning-in-scala/blob/master/src/main/scala/rl/core/StateConversion.scala

I might want to just add a mapObservable.