The existing find_doc_ids function is convenient when the caller
doesn't want to be bothered constructing a term. But when the caller
*does* have the term already, that interface is just wasteful. So we
export a lower-level interface that maps a pre-constructed term to a
document-ID iterators.