chore: save work on radical strokes

Maybe we can solve some missing data if we normalize radicals, since they have so many codepoints with a similar or identical meaning