Move configuration of available datasets from inside the interface to AnnotatedPairs files

[rdnfn]

Currently ff.data.datasets defines the available datasets. Would be nice if less of the dataset information was set inside the code, but rather taken directly from the meta data of annotated pairs files.