Riemannian stochastic optimization methods avoid strict saddle points