updated tests after failing

removed debugging print in generator. Updated training: Agent that have terminated do not receive observation untill all agents terminate.

Added load and save to env (left GridTransitionMap load and save for now)

Added PIL graphics - 5-10 times faster than MPL
move adaptColor into GraphicsLayer
change scatter size arg to s like MPL
move clf, get_cmap up into GraphicsLayer
added boolean "train" arg to ddqn step() to make running faster in player / editor
add plotTreeObs - not finished.
improved pixel scaling in editor (I hope)
Made the agent background steps update a progress bar.

new generators that take agents_handles as extra input and return the agent's initial position, direction and target