Remove the float imprecision from the implementation

The new direct positive-path Hypothesis-based test wants to pass in huge
numbers and the float-based implementation does not do well for these inputs.
Happily, we can do better.