Update seeding with gym and handle Dict Observation Spaces

[saleml]

With recent changes to the gym and gym_minigrid's APIs, the code in this repo doesn't work out of the box. I made two small changes:

• Change how seeding is done in env.py
• Allow more flexibility when creating the ObssPreprocessor for Dict observation spaces

Additionally, I added if __name__ == '__main__' in train.py and evaluate.py, which seems to be necessary for multiprocessing to work on MacOS, as suggested here