Add named type wildcards - #3500

One UX consideration is should ?foo in types emit an error rather than a warning? It's necessary for values, but not for types. If we leave it as a warning, that means you can accidentally leave ?foo in a module inadvertently since we don't persist warnings.

I'd prefer if both type-level and term-level holes behaved the same, so if term holes error then so should type holes.

To add to that - I feel like _ shouldn't ever error or warn, and ?foo should always error.
Now that we have the option to use ?foo, the intent behind using _ becomes "fill in this bit for me" instead of "help me figure this bit out".

Currently we warn if you leave off a type signature for module-level declarations, which is why I preserved the _ warning there. Otherwise, foo = 42 :: _ just becomes a way to not get warnings while still leaving off signature.

I'd prefer if both type-level and term-level holes behaved the same, so if term holes error then so should type holes.

And they should be called kinded holes and report only the kind. 😅

This is ready for review again. I've addressed the UX concerns.

• ?foo always raises a HoleInferredType error. When it's in a type signature, it just does not provide a TypeSearch.
• _ does not warn, unless it's in a module-level type signature, where it continues to throw a WildcardInferredType warning.