When restoring a backup, don't add trees that already exist

I think a hash of window text and URLs+bullets+dividers of the tabs should do for comparison. That would also be helpful for #41.

Start by tracing the backup restoration flow and compare its handling of existing trees with the behavior described here. Review issue #41 for related context, then determine how window text and tab URLs, bullets, and dividers should identify an existing tree. Done means restoring a backup does not add duplicate trees.

Written by the indexing model from the issue text.