The second patch in particular could be cleaned up; there is definitely some
duplication between the two code paths, so opening the database and
creating a query could/should be refactored.

It seems somewhat reasonable to have notmuch_query_search_tags in the
API. I would say just document the limitation of ignoring excludes
when querying for all tags, otherwise that optimization (which is a
huge speed improvement) will mostly be useless.