Physics-informed manifold learning for high-dimensional chaotic dynamics: Vortex-induced instabilities in aerostatic bearings