Fetch a URL and return token-minimized clean text plus a token-savings receipt. Strips nav/scripts/boilerplate so an LLM reads the article, not the page. The receipt counts tokens vs the raw page HTML (typically ~15x fewer, but it ranges from ~1.5x on already-clean pages to 100x+ on script-heavy docs). Two extractors (Defuddle + Readability), body-max selection, so it does not silently drop the article body. Static HTML only — JS-rendered pages may come back partial.