The mix of `anonymiser` and `datasource` didn't quite make sense, as
well as the fact that some of the "datasources" aren't actually
"datasources".

We should split the definition of `datasource` and `repository` types to
fit more what the types are.

This also moves things between the packages for better separation.