Specialize methods on iter::Cloned<I> where I::Item: Copy.

[Stebalien]

Instead of cloning a bunch of copyable types only to drop them (in nth, last, and count), take advantage of RFC 1521 (Copy clone semantics) and don't bother cloning them in the first place (directly call nth, last, and count on the wrapped iterator). If the wrapped iterator optimizes these methods, Cloned now inherits this optimization.

This is a retry of #36791. It was blocked waiting for fixes for #36053 and #36848 to make it into a beta snapshot.

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[Stebalien]

As per the last attempt, cc @rust-lang/libs, @bluss. Also, please wait for travis.

(*edit* apparently I can't call teams :( )

[Mark-Simulacrum]

#1521 seems to be the wrong issue?

[nagisa]

See #28125. I::Item: Copy does not imply Clone has no side effects.

Would be a shame to block such lib-opts on that though, but it has already happened.

[hanna-kruppe]

Since #28125 was closed, RFC 1521 (most likely what @Stebalien meant by "#1521") has been written and accepted, which does allow library code to assume Clone has no side effects for Copy types.

[nagisa]

Aw, right; totally forgot about that.

[Stebalien]

@rkruppe Yes. Sorry about that @Mark-Simulacrum.

[alexcrichton]

The error on Travis seems worrisome?

[Stebalien]

If nothing else, this PR is a good way to find bugs... Closed waiting on #39025.

[Stebalien]

Actually, that's just a dup of #36804.