Followup: Improve Postgres performance

The rewritten table structure of #378 means that nodes are being resolved in another round trip. Though safely cached the round-trips will kill performance.

A follow up on this is to restore some performance by either (a) keeping an encoding on the table as there used to be (subject, predicate, object fields) and parsing them on the Cayley side when they come back from Postgres, or (b) doing an explicit join on the top level.

Or (c), both -- do the parsing in the returning values case, and have a separate optimization iterator that does an explicit join to the nodes table for value comparison.