v1.4.4

Iterative content-block refinement: after scoring candidate containers, the extractor now repeatedly descends into the best nested block (while it keeps >=60% of the parent score). This peels page chrome that large CMS wrappers (Squarespace, Wix, Shopify) carry around the article. Wider boilerplate class exclusions (hero, breadcrumbs, lead magnets, subscribe forms, bylines, menus). 28 tests pass including live Wikipedia/Shopify/Wix/Squarespace URL checks.