Fetch any public web page and return its main content as clean Markdown, ready to feed to an LLM. Boilerplate (nav, ads, footers) stripped. Input: url (public http/https). Returns {url, markdown, word_count}.