Review how the old data is removed from FTS indexes with stable row ids

[westonpace]

The fix in #6216 addresses updating indexes to reflect deleted data when using row addresses but doesn't address the stable row ids case. This issue is to make sure we address the stable row id case.

[westonpace]

At the moment it isn't clear to me this is a problem. In do_create_deletion_mask_row_id we create an allow-list from the row ids to filter search results. This should mean we don't need to prune deleted rows (or fragments) from indices to get correct results (though we may want to do so for better performance and avoiding bloat)