Re: Normalising is REALLY slow

[Chris Warburton]

Implemented "Idea 2":

 - Simplified the process of normalisation.
  - We now require that all locals have already been "prefixed"
    (normalised) as part of the qualification step.
  - This means we don't have to care about locals during normalisation.
  - Normalisation now involves finding the names defined in an
    expression, and replacing those with 'defining-name-1', etc.
 - Before we remove redundancies, we split the given definitions into
   their names and their normal form. This lets us compare normal forms
   directly, then we "plug in" the names later.

Also fiddled to make tests pass.

Some more thoughts to speed things up:

IDEA 3:

We call remove-duplicates for each round of
normed-and-replacements-inner. This is an O(n^2) function. We could
eliminate it if we had mk-output build up the definitions as part of its
output:

 - If the given normal form exists in the hash, we append our names to
   it and we return the definitions list as-is.
 - If the given normal form doesn't exist in the hash, we insert our
   names for it and we cons our definition on to the definitions list.

IDEA 4:

In split-off-names, we call toplevel-names-in and norm, but norm calls
toplevel-names-in. We could eliminate this redundancy if we renamed norm
to norm-names-in, make norm a wrapper which calls norm-names-in with the
result of calling toplevel-names-in, and have split-off-names call
norm-names-in with its existing toplevel-names-in result.