ENH: add benchmark datasets (e.g. BFSongRepository) to `datasets` sub-package

[NickleDave]

in keeping with torchvision structure of package

[NickleDave]

addendum: there is a datasets package but it currently just includes the WindowDataset abstraction (which should probably be deprecated?) and does not currently provide access to any actual benchmark datasets

[NickleDave]

As discussed in #243 https://github.com/vocalpy/vak/issues/243#issuecomment-1200283602

there should be two sub-packages in vak.datasets, datasets.seq and datasets.bbox, that correspond to the two highest-level types of annotations in crowsetta.

Modules for each dataset can then live in these sub-packages