subject, predicate, object and labels are not indexed, thus, using a where clause on these fields for the DELETE statement does a full-table scan. Using *_hash columns instead will use the indexes.
It appears that preparing an INSERT statement on PostgreSQL actually makes it expecting to receive INSERTs and thus, it create some kind of locks for it. If instead, you only send him DELETE statements, it will indefinitely wait for an INSERT and hung.
pruneLocks could create deadlocked transactions on PostgreSQL if multiple locks expired and pruneLocks is called by multiple instances. Also adds some logging.
Inserting packages in a single transaction does not actually buy us anything as we often delete quads during an insertion and thus, Cayley could not use COPY and do a single round-trip. Inserting multiple packages in a single transaction actually creates deadlocks when a transaction tries to insert (A,B) and another one tries to insert (B,A).
