IJBC database will fail on non-Idiap filesystems

Hello, I believe there is an issue with the current implementation of the IJBC database. Currently it downloads a tarball containing pickled precomputed DelayedSample. It works at Idiap because the path to the original directory containing the image data is actually hardcoded, hidden inside the _load function for each DelayedSample.

But there is no mechanism to provide a path to a different image directory, meaning people outside of Idiap won't be able to connect this database wrapper to their own stored version of the database. This is in contrary to the CSVDatabase implementation of other databases, in which the image directory can be provided through the .bobrc config (e.g https://gitlab.idiap.ch/bob/bob.bio.face/-/blob/master/bob/bio/face/database/mobio.py#L68).

Not sure yet how to approach this... I know this pre-pickling approach is important for performance.

ping @tiago.pereira