RLPark Update

[Thomas Degris]

Hello,

some new things in RLPark:

- most importantly, I have changed the Policy interface to fix a design flaw and allow new optimizations. Please note the new method "void update(RealVector x);" that needs to be called to update a policy distribution

- I implemented a discrete (markov) maze problem. A maze can be visualized in Zephyr. Any instance of the interface MazeFunction can be dragged&dropped on top of a maze view to be displayed in the maze.

- new demo in RLPark : QLearning in a Maze: http://rlpark.github.com/doc_codejava_qlearning.html

Next things I am going to work on:

- parallelize the update of HordeUpdatable functions

- javacl in RLPark to integrate Clement's implementation of Horde 

Thomas

PS: an update of Zephyr is required