`train` lists some files twice

The fix for issue #96 (closed) has revealed a series of new issues with train in combination with an imported label file as training data, counting some files twice.

The Info: Pushing file to backend message is showed twice per spine and outline file if these files are data dependencies of a label file.

Similarly the import command writes the data dependencies twice. Calling train on such a file results in a non-blocking message Info: Multiple label files for a same data dependency but eventually crashes later with message JULIA: "no data files found".