While working on PR 11872, it occurred to me that it probably wouldn't be a bad idea to change the `_parsedAnnotations` getter to handle errors individually for each annotation. This way, one broken/corrupt annotation won't prevent the rest of them from being e.g. fetched through the API.