Fetch a PDF by URL and extract its text as JSON. Input: url (public http/https pointing to a PDF). Returns {url, pages, word_count, text}. Bounded to the first 50 pages; 3 MB max.