Specify negative-returning behavior of `abs` method

[klutzy]

Currently ::std::i64::MIN.abs() returns ::std::i64::MIN.
I don't think this is absolutely "wrong", but then at least it should be specified that "abs() method can return negative values".

Currently std::num::abs() says it can return NaN for floats, but no mention for general situation.

[Gankra]

For reference, this is the behavior of Java, C, and C++. But C# yields an error when this is done.

Java in particular documents Math.abs on ints to be:

Returns the absolute value of an int value. If the argument is not negative, the argument is returned. If the argument is negative, the negation of the argument is returned.

Note that if the argument is equal to the value of Integer.MIN_VALUE, the most negative representable int value, the result is that same value, which is negative.

While C# simply states that the input is:

A number that is greater than Int64.MinValue, but less than or equal to Int64.MaxValue.

A third alternative exists in making the output of min be the unsigned version of the input, which should be possible with associated types. You can actually do this today with std::int::MIN.abs() as uint. However I think this solution is generally not well accepted.

[robertg]

Maybe we should document this so there is no confusion?

[klutzy]

Similar issues for rationals:

extern crate num;
use std::i64::MIN;
use num::rational::Ratio;

let a = Ratio::new(MIN, 1);
a.abs() == a; // true

Hmm, I originally thought that we should settle some agreement about negative-returning behaviors and add some notes like "Note: implementation may return negative value if it is impossible to represent correctly".
But now I'm not sure if it is good design that Signed provides abs method at all.