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

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

initial commit