Improvements to estimate_matsize
This MR brings several improvements to the estimate_matsize script, in particular:
- more consistent seeding, so that identical experiments may be run with only a change in randomness
- possibility to use a relation cache, so that a database of relations can be used to test the behaviour of the simulator
- possibility to do random sampling based on the full list of special-q's. This seems nuts. But again, it's one of the ways to check the simulator.
- new
shrink_rels
binary, so that the shrinking strategy can be tested independently of thefake_rels
stuff