Strategy for data set handling in ML notebooks

[mbackenkoehler]

We should have a unified strategy for data sets in notebooks. Sometimes data sets are downloaded in the notebook, sometimes they are in the git directly.