Add model card with evaluation matrix, confusion matrix, and per-class metrics

Upload fine-tuned LLaMA-3-8B CEFR model with confusion matrix

initial commit