Tests pass, except Miri (see rust-lang/miri#950). But I'll address other distributions before merging.

The Cauchy test is generating a very slightly different value on OSX and Windows to that on Linux. 😞

Presumably this is because the tan function is provided by system libraries, which have varying degrees of precision. I don't see an alternative to working around this issue. (Maybe someday Rust could have its own implementations of trig/math functions?)

The above mentioned issue has been worked around; the only failing test is Miri (unrelated).

Miri should be good again now, sorry for that.
Hopefully, we'll get rust-lang/rust#60301 resolved soon-ish, which should avoid such broken Miri components being shipped. But my first attempt at solving that issue just failed spectacularly. :/