arxiv SVTR: Scene Text Recognition with a Single Visual Model