Add Grapheme#each_char and #each_byte

[straight-shoota]

This patch adds two iteration methods to String::Grapheme. They delegate to the respective implementations of Char and String.

[asterite]

I'm starting to think that instead of Grapheme, we should have maybe returned a String as each grapheme, even if this is not the most efficient thing. Maybe we could have an array of strings for ascii chars and use those so we don't create chars for those. The thing is, it looks like Grapheme should behave exactly the same as a String. In fact, in Elixir there's no grapheme concept: you always get strings back.

[straight-shoota]

A table for ASCII characters won't help for non-ASCII characters. But I agree that the data format for graphemes is not entirely finalized. So I'd suggest to continue this discussion in a separate issue. And for now we can add the Experimental annotation to the grapheme API, so we can refactor it in one of the next releases if we want to (=> #11611).

[beta-ziliani]

Perhaps a bit obsessive, but IMO it's missing two cases: an empty string, and (a) grapheme(s) with more than one codepoint.

[straight-shoota]

A grapheme consisting of an empty string should be impossible to acquire. I don't think it's necessary to test that. I would see that as an invalid state with undefined behaviour.
These two highlighted examples have more than one codepoint. Maybe it's a bit confusing that they are not actual graphemes. I chose sequences of codepoints that don't form an actual grapheme for demonstration purposes. This is unrelated to the grapheme breaker algorithm, so it doesn't matter. Let me know if you think I should use actual graphemes instead.

[Sija]

These are copy-pasted from above, testing each_char