Supporting a selector filter is a really good idea to avoid boilerplate. If you submit a pull request to that effect, I’ll gladly merge it (try to include docs too).
I agree that the need to define your own parsers for all nodes when you add an attribute is somewhat awkward, but I do feel that handling this abstraction in your own code (a wrapper that you reuse across parser functions) is no more complex than a solution that involves extending the default parsing strategy would be, and I would like, as much as possible, to avoid complexity in the core.