RL Bench

It is now possible to use 97 of the 101 RL Bench environments (place_cups, remove_cups, put_plate_in_colored_dish_rack, pour_from_cup_to_cup have bugs). The main link between our framework and RL Bench is the RLBenchWrapper, which extends the RLBenchEnv(gym.Env) with methods needed for goal-conditioned RL.

I ran all environments for one epoch with 2 episodes. There may be bugs that only occur later.

reach_target-state-v0 has already been trained with HAC (2xsacvg) and achieved a test/success_rate of 0.8 at epoch 11 (each epoch had 2000 steps).

This MR also contains the Coppelia Reacher environment.