Riemannian Flow Matching Policy for Robot Motion Learning