Towards Early Prediction of Self-Supervised Speech Model Performance