Conditionally parse comment nodes

These should be lazily parsed. Also, when someone forgets all child extended comments, it shouldn’t parse them on the next manipulation (perhaps the containers should have a `commentCount` property or something... will figure it out when I’m not sleepy).

Start by tracing how comment nodes are parsed and how container manipulation triggers parsing; the issue does not name specific files or tests. Define the lazy-parsing behavior and verify that containers with no child extended comments do not parse them again during later manipulation.

Written by the indexing model from the issue text.