Fix bug in counter example pruning script

[nomeata]

Strangely, even after updating implications.json to the latest version, the unknown count increases from from 2726304 to 2734065. Not sure what’s going on, but as this is definitely fixing one (among more than one?) bug, probably still worth merging.

This does not update the All4x4Magmas files, because I’m confused how they should be generated these days, see
https://leanprover.zulipchat.com/#narrow/stream/458659-Equational/topic/4.C3.974.20magmas/near/474767268

[zaklogician]

The Refutation*.lean files checked into the current main branch are generated using the generate_5x5_refutations.sh bash script from the file plan-5x5.txt (itself generated from other files by a heap of tools starting with Mace4, although a copy is checked in so we don't have to rerun it every time).

The bash script itself calls out to a Haskell tool, gen-refutations in src/finite_magma_tools which parses data/plan-5x5.txt and emits the Lean code directly.

However, gen-refutations does not read implications.json and does not do pruning - it's a throwaway tool, meant to eventually be replaced by an updated generate_lean.py which can handle something like plan-5x5.txt. I made an attempt at doing precisely this (see the repo linked in #265) but bounced back when the implication counts didn't match up.

I think this bug might have been the cause, so I'll revisit #265 by rebasing my port over this once this is merged.