bsuite environment implementation

[Robert Tjarko Lange]

Implement the set of classic bsuite environments:

• catch.py
• bandit.py
• deep_sea.py
• discounting_chain.py
• memory_chain.py
• mnist.py
• umbrella_chain.py
• cartpole.py - Check if this is different from OpenAI gym version
• mountain_car.py - Check if this is different from OpenAI gym version

[Robert Tjarko Lange]

CartPole seems to use different hyperparameters/clipping. Leave both gym environments for know in gym style.