Setup and Modified Tomato Watering Env

• Added a setup.py for easier installations
• Created a modified tomato watering environment that satisfies the definition of a Corrupt Reward Markov Decision Process
• Allow for newer numpy versions: More recent numpy versions may break things, but current tensorflow binaries from pip are compiled with numpy>=v1.15.4.