Introduced "LLM as a judge" pattern for better book metadata. Now the site intelligently decides which book "edition" to display from Open Library. Rather than using some suboptimal heuristic that consistently puts out suboptimal (or straight up non-english) descriptions, we now use LLM-as-Judge Pattern with better candidate collection to pick the best from all available descriptions from OpenLibrary and Google Books. Only if those result none, do we generate them from scratch.