Improve usability of abelian_closure(QQ), e.g. for elements representing real values allow comparisons and conversion to Float64

[ThomasBreuer]

resolves #3976

[ThomasBreuer]

The way how n-th roots of unity are printed is irritating:
On the one hand, the documentation says that the default is z(n).
On the other hand, the docstrings show zeta(n).

[thofma]

If the documentation is lying, maybe we could change it? To make it somewhat consistent.

[fingolfin]

Looks great to me, thank you!

Two minor nitpicks...