longTableOptimize

A user showed in an example that the platypus Table class was fairly abysmal 
even with the longTableOptimize config as True.

Seems we had a quadratic issue we calculated row heights and then split with the 
original argument row heights. If the row height calcs are non-trivial the 
result is poor.

I have tried to fix this case by re-use of already computed values. All the 
tests seem OK and the huge 10000 row table that was causing the trouble seems to 
go much faster.

This change was committed in rev b6902985cf64 (changeset 4266).

If anyone has an unwanted change caused by this please let me know.
-- 
Robin Becker