Remove Atomic's _owned methods

[jeehoonkang]

Closes #30

[ghost]

Nice! I really like how this panned out. :)

[ghost]

This looks kinda messy at the moment, but with in-band lifetimes (see here) and impl Trait in argument position (see here), this signature will be simplified to:

pub fn compare_and_set<P: Pointer<T>>(
    &self,
    current: Shared<T>,
    new: P,
    ord: impl CompareAndSetOrdering,
    _: &'g Guard,
) -> Result<Shared<'g, T>, (Shared<'g, T>, P)>

[Vtec234]

The return type stands out as really sigil-heavy and could potentially be changed to Result<Shared<'g, T>, CasError<'g, T, P>> with fields like curr and new in CasError or even entirely to CasResult<'g, T, P>. That said, the current interface uses standard types and is understandable without looking up more custom types, so i'm not so sure. The linked RFCs will definitely improve things.

[ghost]

I'm in favor of CompareAndSetError<'g, T, P> because it's easy to mistake the tuple (previous, new) for (new, previous).