Minor refactoring and performance features.

The refactoring is mostly confined to the SQL 'ORDER BY' generator. I
had intended to make it more powerful, but that turned out to be more
work than I expected. Instead, I'll settle for clearer code.

The performance features aren't likely to be useful unless you're doing
heavy analysis or are using nfldb to power a web site. Briefly, several
`fill` functions have been added to nfldb's entity types that will, for
example, fill in a list of plays with drive and game data using as few
queries as possible. Currently, this will fall over with >5000 items
because of PostgreSQL limits, but future work will fix this using some
sort of batching mechanism.