finish gym interface

[Armandpl]

• make sure sim timing and robot timing are the same
  • e.g if we call step on the real robot we get back the angle readings at t
  • make it so its the same in sim
• handle reset
  • wait for it to stop moving
    • handle the case were it stopped off from 0 which probably means the encoder slipped -> eh just reset every time
  • reset motor position to zero
• make sim follow the same interface as Robot in robot.py THEN wrap in gym env
• vectorize training
• clean up the way we instantiate wrappers
• check how to seed env now that we're supposed to pass seed to reset? should we pass the seed to sb3? why are we seeding on reset even?
• make it so we use DummyVecEnv and not SubprocVecEnv when training on the real robot
• fix video saving. currently it fails and only records the first frame.
  • only on the remote machine?
• fix model names when saving loading. use the same name for every algo
• finish choosing how to setup state_space #56
• scale viewport width scale based on the max/min theta angle. e.g -width/2 = -min_theta
  • use the safety limits? or the obs limits? or just -pi/pi and wrap? I think that makes more sense
• display degrees on the cartpole track
• add pytest to CI
  • can we mock the robot to test the real env?
• quickfix for match case not available in python3.9 #57 to pass tests and lint