This new format of CSV databases fixes the following issues with the old format:
There is a clear distinction between subject_id (which is the identity of the person behind a Sample) and template_id (which is the unique identifier of a biometric template, for both enroll and probe SampleSets)
Implements sparse scoring through a separate for_matching.csv file. In the old format, this was possible through hacks only.
Removes the SampleLoaders
The implementation is based on the one in bob.pipelines