To allow for a more compact provenance index, we should try running the provenance algorithm on the subset of revisions that match a release.

This would be:

• release objects (eventually pointing at a directory)
• revisions in a snapshot matching a known "tag" identifier (refs/tags/*)