[developers] characterization

[Ben Waldron]

I've checked in a number of fixes to the old characterization code in 
the LKB. Specifically:

- The active parser now outputs correct characterizations (for 
rule-contributed semantics);
- The code is less inefficient: the active parser runs approx 6% slower 
when characterization is enabled (was 20%), and uses approx 4% more 
memory (interestingly, there is a net improvement in memory usage for 
longer sentences).

(New functions set-characterization-tdfs-within-unification-context and 
set-characterization-tdfs are used in place of the old 
set-characterization.)

To enable characterization *characterize-p* must be non-null.

- Ben