training script modification

[samuelmyoussef]

Modified the rl.py script:

• Removed algorithms from stable baselines (SB2)
• Added the remaining algorithms from stable baselines (SB3): A2C, DQN, TD3, DDPG
• Added all the available environments in SofaGym
• Added command line arguments to choose the env, algo, and training parameters
• Added the choice to continue training from a previously trained model
• Added Tensorboard logging

Added a documentation part to briefly explain the script usage.

[samuelmyoussef]

I made modifications according to @alxbilger recommendations:

• Implemented classes for the RL agent to separate them according to the framework used:
  • SofaTestAgent: to create an agent only to test the env
  • SofaBaseAgent: the base class for the training classes
  • SB3Agent: to train the agent using Stable Baselines3
  • RlberryAgent: to train the agent using rlberry
• Added all model hyperparameters and training parameters into yaml files to be edited directly instead of parsing command line arguments.
• The rl.py script could now be used to create and train an RL agent by specifying the wanted environment, the algorithm, and the framework to use (SB3 or rlberry for now), and the interface is the same whichever the framework chosen.