doc: clarify edge cases when converting rational to float (#45220)
We state that for any integral `a` and `b`, the expression
`isequal(float(a//b), a/b)` is true unless `a` and `b` are zero.
This is confusing because there are two such cases, which both require
only one of `a` or `b` to be zero. The first case is the division
by zero and the second case uses a negative divisor to make float
division evaluate to -0.0 which has no equivalent rational number:
julia> isequal(float(0//-1), 0/-1)
false
Clarify the conditions of the exceptional cases.