TODO:

for Apple site:
make duplicate sentence detector
add common missing lexemes

FIX gen-counter overflow
	- 32-bit counter is pretty big though
	- 32-byte struct dag, 1GB RAM limit currently per sentence
	- so, 32kdags per sentence max (actually much less due to arcs)
	- figure 1/3rd of unifications fail
	- so conceivably 100k unifications per sentence!
	- limits us to 40k sentences per run :(
	- appleout is in danger, then
	- realistically though, only about 2kdags per sentence are produced
	- so maybe we can do 400k sentences per run

make equivalent_dg actually work...

multi-word stems for lexical entries
make ambiguity packing work with semantics (and record packing)
preprocessing fsr?
statistical parse selection?

OPTIMIZATIONS:

find a safe way to not copy types for extra constraints

use different quickchecks for different rules

find a way to force struct dg's to align to 32 bytes,
	so they fit one per cacheline nicely
find a way to force struct darcs to align to 8 bytes
smart ordering of agenda

hyperactive parsing -- for certain rules only