Text this: Assessing how accurately large language models encode and apply the common European framework of reference for languages