Re: Normalising is REALLY slow

[Chris Warburton]

raco profile shows that reps-insert-rep-acc is taking most of the time:

 [47] 122422(78.7%) 122422(78.7%) reps-insert-rep-acc .../lib/replacements.rkt:168:0

We need some ideas to speed this up.

At the moment, we're doing the following loop:

 - Start with an empty result list
 - For each rep, loop through the result list
  - For each entry, see if it's disjoint from the rep
   - If so, go to the next entry
   - If not, merge this entry with this rep, and start again

Starting again seems a bit wasteful; what extra invariants are there we
can use?

We could just "pull out" the entry, e.g. using a zipper, and restart
that particular iteration; since we know the other reps are still
disjoint from each other. That way, rather than building up a correct
result via backtracking, we'd be accumulating and collapsing as we go.