It turns out that a registry doesn't build a configurable correctly as a subtask – it is missing some important information, including a log. Rather than put up with this, I plan to ditch the registry and use our standard subtask solution. It standardizes the code and presents one preferred way of doing things instead of two different ways.

Done. The star selectors are now tasks and so have task documentation, but no complete examples due to the lack of easily available calexp and associated star catalogs. Once those become vailable then simple examples will be very easy.

I've linked DM-1102 here to at least start to capture that need for better unit test data.

There are a few things about this code that make me uneasy. I have listed them on RFC-154.

If you have time, Jim, I'd appreciate your reviewing this code (or sanity-checking it if you don't have time for a full review). The registry is gone, which I don't expect you to be happy about, but flagging stars is supported. I am happy to take a final cleanup pass on commits if you like, once we are sure this is the final approach. They commits are fairly clean now (I hope) but show some evolution of the design.

Looks great overall. I have a few comments on a couple of the PRs, but nothing major.

Thank you Jim Bosch. I made all requested changes (except I will ask you to file the ticket about making code in one star selector more visible to others, since I didn't understand the use case). I will merge after a successful Jenkins run.