v0.1.6

Repair release. Every fix here came out of using nakkas the way an AI actually uses it.

Fixed

• Background ignored the viewBox origin. With a centered viewBox like -100 -100 200 200 the background covered only one quadrant. It now anchors to the viewBox rect.
• Generic font families rendered with the wrong font. monospace, sans-serif, and serif fell back to an arbitrary system font in preview and PNG save because resvg's family mapping options are non-functional. Generic keywords are now resolved to real fonts before rasterization, using fontconfig's own mapping on Linux and platform defaults on macOS and Windows. The preview finally matches what a browser shows.
• Validation errors were blind. elements.1: Invalid input became elements.1.content: Required, with a field cheat sheet for the failing element type appended to the error.
• Docker text rendered as blank glyphs. The image now installs fontconfig and DejaVu fonts.

Improved

• render_svg tool description documents the field names that differ from raw SVG: text uses content, textPath uses pathId plus text, groups use children, pattern groups take a single child.
• README rewritten around the preview loop, with honest font guidance and an AI-readable summary. llms.txt and llms-full.txt synced.
• server.json version tracking fixed; the MCP Registry entry had gone stale at 0.1.4.

296 tests passing.