Dataset queries done using fixed IDs are 2-3 times faster than queries
done using the where keyword, a significant savings when querying the
central repo. However, queries against a set of IDs cannot be converted
to this form and must continue using where.

The comment justifies why we can't filter by validity range, but we
have been doing so for a while.

Butler registry queries against all dataset types are an order of
magnitude less efficient than an equivalent set of queries that
enumerates all possible types.

Creating a function that returns the types makes it possible to replace
the hardcoded list with something more flexible later.

This still forces us to load refcats the pipelines don't use, but at
least prevents us from having to manually curate a list of known types.