A Survey of Geometric Optimization for Deep Learning: From Euclidean Space to Riemannian Manifold