lark-performance-tests

This is a mess. Was created for lark PR #949.

These scripts made sense to me at the time. Not sure how about now.

In general:

• performance_tests.py uses a second implementation of the rule generation to tests a lot of different scenarios on how splitting up the N into as and bs performs.
• performance_tests_2.py varied the actual Thresholds inside of lark to check for the actual effects on the implementation.
• best_factors.py just tests which 'factorizations' of as and bs gets the best 'factor_sums'.