Implement Float equality using ULPs ourselves
We're using the "float-cmp" crate to implement float equality using ULPs (https://randomascii.wordpress.com/2012/02/25/comparing-floating-point-numbers-2012-edition/), mostly so you can compare simple floats more reliably and sort a list of floats. The implementation of this is simple enough that we can just implement it ourselves instead of introducing a dependency. This would let us remove the "float-cmp" direct dependency, the indirect dependency on "num-traits", and the built-time dependency "autocfg".