comprehensible data preparation scripts

[mateuszwyszynski]

The current way the data scripts (like data/prepare_data.py) work is very messy. Among many problems:

• there are hard-coded paths from the original paper author's machine
• the data files which are nested more than one level are not automatically read by the data/prepare_data.py script. One has to manually move these files. This should be automated

Overall the current data preparation process is extremely unclear. There are unanswered questions like:

• which datasets from the AMASS should be downloaded
• how to train a model using a smaller subset of datasets
• after downloading datasets, how exactly the data should be organised for the scripts to work