Add GenericPatternItemULE::as_pattern_item_ule

[sffc]

Smaller change pulled off from #4415

This includes a bit of unsafe code and takes advantage of the fact that currently GenericPatternItemULE and PatternItemULE have the same repr if they are representing a code point. I added a debug assertion to ensure that this invariant remains true.

[Manishearth]

code lgtm

I'd like the repr docs for both of them to include a mention of this soft invariant ("if changing the repr, consider making sure this transmute is still possible") so that when we ever go to change this there are early warnings.

r=me with that, feel free to merge beforehand

[zbraniecki]

ain't gonna block on it, but the (ab)use of Result to indicate enum is unintuitive for me.
Result::Err indicates not just the value but also a misalignment with state expectations. In this case Err is a perfectly expected and reasonable state.

[sffc]

ain't gonna block on it, but the (ab)use of Result to indicate enum is unintuitive for me. Result::Err indicates not just the value but also a misalignment with state expectations. In this case Err is a perfectly expected and reasonable state.

I argue that it is a valid use because the fn as_pattern_item_ule either succeeds or fails at the conversion. I suppose try_as_pattern_item_ule might be a more appropriate name. But, this is a private fn.