Vision language models frequently enter repetitive failure loops when encountering repetitive characters such as rows of periods used as spacing in PDFs