Use UnvalidatedStr for casemap unfold keys

[Manishearth]

We have two other strings in casemap on data that isn't hit often: the exceptions data and the unfold values are also strs.

@sffc @robertbastian what do you think of making those UnvalidatedStrs as well? We could GIGO when they're not. The nice thing is that people won't pay upfront costs about data that is not used often.

Unfold data is not too big, exceptions data is kinda big but also not all of it is rare to hit.

The main issue is that in the exception data multiple calls to get_char_slot() etc would have to reparse the string. But the string is likely to be small anyway.

[robertbastian]

I don't think we should make values unvalidated until we know that it's actually a performance improvement. We don't know much about real usage patterns.

[Manishearth]

That makes sense to me.

(Merging this PR to get the change for keys in, which is unlikely to be controversial; for values I'm happy to continue discussion if others feel like we should do something there)

[sffc]

For now let's use UnvalidatedStr in zeromap keys but not zeromap values assuming that the values will be used as strs.

[Manishearth]

(The values here are logically [char] stored as UTF8, not strings. But the same argument applies)