Improve RawIter re-usability

[jonhoo]

Creating a new iterator for a table is occasionally expensive,
especially if a suffix of the table (since we iterate "backwards") is
sparsely populated. This PR adds methods that allow re-using an existing
RawIter in more situations.

Replaces #167.

[Amanieu]

I like this much better!

[jonhoo]

Glad to hear it! I have now also implemented the necessary changes in griddle, and it seems to be sufficient: jonhoo/griddle#12. The Clone implementation isn't as pretty as I'd like it to be, but it's not too bad.

The griddle test suite also passes, including a number of additional iterations of the quickcheck tests I've run locally, so I think this is ready now 🎉

[jonhoo]

One really awkward side-effect of this (and on jonhoo/griddle#12 in particular) is that griddle's HashMap now cannot implement Clone without K: Hash, S: BuildHasher bounds, since it must perform inserts. This, in turn, means that derive(Clone) will no longer work 😢 griddle::raw::RawTable can't be Clone at all, since it requires a hasher.

[Amanieu]

Clone probably should depend on K: Hash, S: BuildHasher. The only reason hashbrown doesn't is because it needs to be backward compatible with the libstd HashMap which doesn't have those bounds on its Clone impl.

[jonhoo]

Is there anything more you'd like to see added to this PR? I've tried to document as I went, and have also added the appropriate "resume from RawIter" methods for drain and into_iter.

[Amanieu]

Thanks!