Learning to Land Anywhere: Transferable Generative Models for Aircraft Trajectories