Renamed notebooks to reflect conferences,
Empty placeholder packages (rl, notebook_modules) for future refactoring,
Hyperparameter optimization notebook + dependency,
meta.py: library ranking, pruning by clustering, new distance metrics,
ppo.py: Added step-wise and interval callbacks,
systems/tanks.py: Higher resistances to make open-valves lag, new reward function with normalized components, bugfix with median tank

More: turbulence parameter is float, takeoff simulation to get motors running in the beginning.

systems.quadcopter: Refactoring, yaw control optional, deterministic resets, trialing new reward, no ground level for simulation

TODO: Finish reward formulation, episide end. Verify env reset propagates to controller and quadcopter

Need to look into box2d engine to set env params. Also make module-global parameters local to class so different environments can have different parameters

Implemented 2nd order, 1st order, reptile approximations to meta-update;
Possible bugfix in tanks system;
Implemented Standard MAML for comparison;
Switched to Jensen-Shannon distance for pruning library of policies;
Added ranking function for selecting policies;
Added seeding for reproducibility (some left);

Bugfix with cache to episode conversion in utils.py. Done flags can either omit terminal state or initial state of next episode.