Instead of reading the OPF xml file twice, once to get the id and a
second time to resolve the id to an href, collect both in one go and
stop when the cover href is found.

As a side effect this enables the fallback heuristic when there is an
id but the href can not be resolved.