v0.25.2

Fixed

• Word clouds no longer collapse to their title in embeds (Obsidian, remark/markdown, web embeds). Word-cloud words are positioned with transform="translate()" and have no x/y attributes, so the embed normalizer's string bbox estimator measured only the title and zoomed that fragment to fill the frame. Tightening is now rejected when the measured box covers less than half the renderer's canvas, keeping the correct full-canvas viewBox. Extends the embed-clipping fix in 0.25.1; verified in WebKit.