Audio-JEPA: Joint-Embedding Predictive Architecture for Audio Representation Learning