In the UnixFS representation of the dataset directories are nodes and have a CID, and we know in principle all the linked files below them, so is it worth recording the CIDs of directory nodes to allow clients to grab 'everything under here'?
Different clients work differently in this regard but at leat we'd give them the opportunity.