Gym environments and stable-baselines integration for RL

[saihv]

About

OpenAI Gym wrapper for AirSim, and sample drone/car envs for reinforcement learning. Contains DQN examples that will replace the previous hand crafted DQN code. Integrated with stable baselines 3 (PyTorch based) for easy RL training with standard algorithm implementations.

Testing

This PR requires Pytorch to be installed (https://pytorch.org/get-started/locally/), along with the following specific dependencies.

pip install gym tensorboard stable-baselines3

DQN car: Open Neighborhood binary, and run python dqn_car.py
DQN drone: Open LandscapeMountains binary (with powerlines) and run python dqn_drone.py

[zimmy87]

Tested this locally and it works as expected. Please ignore the two failing Azure pipeline checks, those are failing due to an unrelated issue.