Deprecate Float::abs_sub

[frewsxcv]

{f32,f64}::abs_sub is misleading and has been deprecated for a couple years. Maybe it should be deprecated in the Float trait too?

[cuviper]

In num-traits we have this in Float, Real, Signed, plus the free function num_traits::sign::abs_sub.

I'm ambivalent. It is surely a weird function, but it's also harmless to leave it alone. It would be more meaningful to wholly remove it (as a breaking change), so trait implementors don't need to bother with it, but deprecation doesn't affect this.

[CAD97]

You could provide a default implementation equivalent to (lhs - rhs).max(Zero::zero()) to allow implementors to pretend it doesn't exist.