Multilink fixes and optimizations: - Optimisation: Late evaluation of Multilinks (only in rdbms backends): previously we materialized each multilink in a Node -- this creates an SQL query for each multilink (e.g. 'files' and 'messages' for each line in the issue index display) -- even if the multilinks aren't displayed. Now we compute multilinks only if they're accessed (and keep them cached). - Add a filter_iter similar to the existing filter call. This feature is considered experimental. This is currently not used in the web-interface but passes all tests for the filter call except sorting by Multilinks (which isn't supported by SQL and isn't a sane concept anyway). When using filter_iter instead of filter this saves a *lot* of SQL queries: Filter returns only the IDs of Nodes in the database, the additional content of a Node has to be fetched in a separate SQL call. The new filter_iter also returns the IDs of Nodes (one by one, it's an iterator) but pre-seeds the cache with the content of the Node. The information needed for seeding the cache is retrieved in the same SQL query as the ids.