Handle large (1000+ page) documents better

Currently, we generate the DOM objects for all pages at init, but we could be smarter about loading only a small subset of these pages, then adding more (and removing some) as necessary.

Start by tracing the initialization path that creates DOM objects for every page and identify where page loading and removal can be controlled. Compare behavior on a document exceeding 1,000 pages, then define completion as keeping only a small needed subset in the DOM while adding and removing pages as navigation requires.

Written by the indexing model from the issue text.