Special cases for +/- 0 around branch cuts

[asmeurer]

We should notate that +/- 0 should give the corresponding side of the branch on a branch cut. See the discussion at #514 (comment). For cuts on the real axis this should more or less be covered by the text about conj(f(x)) == f(conj(x)), but there are also functions with cuts on the imaginary axis (like asinh).