More specialization tests

[Philippe-Cholet]

• To do the next point, test_specializations need clonable iterators which helped find iterators that should be clonable but were not yet. 3 very small commits.
• I do not want to add specialization tests one by one in future pull requests so this is mostly done! 1 big but straightforward commit.
• For two tests I previously wrote, I now discard inputs (in the quickcheck-way) instead of truncating them. 1 small commit.

Now about mostly done... Not all of our iterators are tested here:

• group_by
• chunks
• tee
• rciter
• test_specializations need clonable iterators but &mut iterator is not clonable, Therefore, we can't clone PeekingTakeWhile and TakeWhileRef. We need to find a way to test them too.

Finally a bug... multi_cartesian_product is not tested here because specializations of last and count FAIL sometimes, and I don't want this pull request to be blocked by it! It requires further investigation and will have a dedicated pull request.
EDIT: Apparently, the iterator is not fused even when all child iterators are, and this was the intended behavior, that's confusing.

Later EDITs: While making a bunch of specialization benchmarks, I noticed that I forgot some and some here:

• repeat_n
• TupleBuffer (buffer for a finished .tuples())
• MapForGrouping used for GroupingMapBy but it's strictly internal to the library, a method of GroupingMapBy should be tested to test the specialization of future MapForGrouping::fold.
• unfold, iterate: infinite iterators so e.g. test last would not finish.

[jswrenn]

This looks good to me. You can go ahead and merge this as-is, or you can add a test for multi_cartesian_product and mark it as #[ignore] (so we don't forget about it, but it also doesn't fail CI).