#1366 - Approximate equality test for Rationals

[schillic]

Closes #1366.

The last commit is unrelated, but I found it weird that we have different (and actually inconsistent) behavior for Rationals. I would actually argue that we should also remove the default for _leq for the same reason, but I did not have a closer look at it.

[mforets]

Rationals can and should use exact comparisons; what i suggested here to share the interface, is to just let isapproxzero(x::Real, ...) handle it (but don't remove it!).

[schillic]

Yes, but the other implementation uses exact comparison, so I argue that we can use the float method in all cases (which simplifies things). Furthermode, if one wants to change the tolerance for new number types, one can simply override ABSZTOL and it works automatically. This is not the case with the removed method.

[mforets]

Oh, true, with ABSZTOL(N::Type{Rational{INNER}}) where {INNER} = zero(N) even abs(x) <= ztol has the same behavior as checking that x is zero.