Hidden Convexity of Wasserstein GANs: Interpretable Generative Models with Closed-Form Solutions