Clarify: all applicators collect annotations, including "contains"

[handrews]

As noted in #766, we need to emphasize how applicators and annotation collection works, in particular that an applicator like contains does collect annotations, which means that when annotation collection is enabled, you cannot short-circuit the evaluation. I think we note this with anyOf but not contains. Those are the only two applicators that combine validation results with OR and could therefore be short-circuited when not collecting annotations.

[gregsdennis]

All annotations? What about not? It's been made clear that it will never return annotations due to its nature.

Otherwise I think this is a good idea.

[handrews]

@gregsdennis annotations are still dropped whenever validation fails (and it's impossible to evaluate not without either failing the subschema, before not passes, or passing the subschema, in which case not inverts the validation result to fail)- this is a note to myself to fix a thing, don't treat it as spec. NOTHING IS CHANGING HERE.

[handrews]

@gregsdennis ok apparently I'm a lying lier who lies because contains currently says it does not collect annotations 😮

I'm 99% sure that is because we didn't have any sort of output format when I first wrote the spec for that keyword, and it wasn't clear how to make use of them if they were collected. That is no longer the case, so I'm writing the PR to change it to collect annotations and be a "normal" applicator.

We can debate whether that is the right option on the PR 😄

[karenetheridge]

clerical error? This issue is in milestone draft-09, but the PR that closed it is in milestone draft-08 (which maybe should be subtitled ".. (2019-09)").