Stack models + dataloader

[araffin]

• Add support for splitting the representation learning (e.g. one dimension for AE, two dimensions for inverse + forward)

• Update dataloader (simplify it using joblib)

• moved different files to evaluation/ folder

• Add a script for predicting the states using a trained model + compute stats

• Remove unused function in learner + add a method to load a trained model

• SRL training is now fully deterministic

• [x] Clean up model splitting + add documentation

• [x] Speed up new dataloader

• [x] Add support for model splitting in RL

[araffin]

The only thing missing: speeding up the refactored dataloader

[araffin]

Missing before merging: fix issues with multi-view cam (data loader) + add documentation to new data loader

[araffin]

Tests: 8 passed in 676.16 seconds 82% of coverage