During the Verifpal workshop today at Eurocrypt 2022, a bug was
spotted by Oskar Goldhahn (NTNU) which led to two very similar models
(included here as `examples/test/ordering_a.vp` and
`examples/test/ordering_b.vp`) outputting very different results. After
the workshop, I did some debugging and figured out that some positively
hare-brained "performance optimization" I had implemented roughly a year
ago seriously reduced analysis precision while basically adding no
performance benefit that I can subsequently measure. This nitwitted
and positively idiotic "optimization" is happily removed here.

Thanks to Oskar for spotting this bug. Happy to report that an
ill-guided optimization was to blame, and nothing more fundamental to
Verifpal's analysis logic.

Equivalence patch

See merge request !12