Streamline Routine quite a bit

Including the sort_dispatchees logic.  This causes "make test" to drop
from 202 CPU seconds -> 195 CPU seconds, and lose a few seconds wallclock
on "make spectest", as well as installing core modules < 8 seconds
wallclock.

In other words, people may notice this in compilation time, and possibly
in runtime as well.

Should probably also look into changing the %info and %graph hashes into
proper objects for performance.