Since it's a static site, splitting the content across multiple pages and still being searchable would mean you'd need to push the whole search index down to the client and have the search operate locally. Not that it couldn't be done, and I can think of a few optimizations, but for the "very large docs" use case that's called out in the readme, it might still amount to a lot of data
on top of the "a lot of data" that constitutes the docs themselves as rendered. Relying on browsers' built-in text search makes a lot of sense, especially to start with; I'd likewise wait to see a very strong use case before building out my own search to supersede it.
I don't know how I feel about the way they're managing the URL fragment. The way they're calling pushState, it adds history entries while scrolling, and that makes the back button behave nonintuitively; I think I'd rather see history entries created only when I click a header, or otherwise take some specific action to demonstrate the intent. That makes the back behavior easier to reason about, and also preserves the easy deep linkability that they are, imo rightly, going for.