A three-dimensional multi-modal foundation model for optical coherence tomography