Control over buffering of comments

Currently comments are buffered entirely for the comment handler.

Documents with pathologically long comments can hit the memory limit, and the user has no way of preventing that, other than not using comment handlers.

It's wasteful and unnecessary if the handler intends to remove comments.

I think we could add some control for it, e.g. add a configurable maximum length of comment to buffer. If the comment is longer than this, then discard the entire comment (with 0 buffer len acting as automatic comment removal). Another option could be to split the comment and handle it in fragments, but that would be as complicated as text chunks (#256).

Read the comment-handler buffering path and compare the text-chunks work referenced in #256. Decide whether the feature should discard comments beyond a configurable maximum, treat zero as automatic removal, or split comments into fragments; done means the chosen behavior prevents unbounded comment buffering while preserving handler control.

Written by the indexing model from the issue text.