I added some features on CartPole

I wanted to do a first test on PPO to see what works and what doesn't.

Now that I found good parameters for everything but the number of neurons in the actor and the critic, I wanted to isolate the number of neurons and see what happens.

For a first sweep, I tried DDPG on Pendulum-v1, as only fixed parameters the number of neurons in the two neural networks (400, 300). I discovered later that the parameters were not correctly initialized in the agent, so these are runs of the same agent over and over.

For the second sweep, I tried DDPG on Pendulum-v1, as only fixed parameters the number of neurons in the two neural networks (400, 300). No bugs, this time.