Add args to configure the training runtime + save training information and results.

[hidal00p]

Args

• Expose argparse configuration.
• Add algorithm selection.
• ~number of training episodes.
• max episode length.
• n cpus

Save format

• Metadata should be human readable (a pretty compact .txt information).
• Model is saved in a binary format as per stable baselines implementation.
• This should enable a loadable scenario.

[hidal00p]

This would probably be my final task apart from potential reward function tweaking.
I would like some easy API to load the whole env i.e.

• Trajectory
• Obstacle locations
• Model
• Eyes