Add model card with exact and within-1 confusion matrices and per-class metrics for non-fine-tuned Mistral evaluation

Upload non-fine-tuned Mistral-7B CEFR evaluation with exact and within-1 confusion matrices

initial commit