Accelerating the discovery of steady-states of planetary interior dynamics with machine learning