An arc length computation test fails with insufficient precision on Travis-CI for Python 3.8

[konstantint]

See https://app.travis-ci.com/github/konstantint/matplotlib-venn/jobs/616934442

[skupr-anaconda]

Is this issue caused by numpy 1.24 (?) and python 3.8? We got the same error only on linux-64, see AnacondaRecipes/matplotlib-venn-feedstock#2 (comment)

[konstantint]

I do not have a very good idea of what the real problem here was, but it does seem that different numpy versions on different python versions can produce different types of objects (e.g. sometimes a float, sometimes a np.float. Most confuslingly - sometimes a bool, sometimes a thing like np.True_). I unified some of the code to always return one type in c742161.

As Travis stopped my builds because of some pricing changes I don't really know whether it would still fail on Travis in some configuration, at least I cannot seem to reproduce any failures on any versions locally.