Two-Stage Learning of Stabilizing Neural Controllers via Zubov Sampling and Iterative Domain Expansion