Fast Text-to-Audio Generation with Adversarial Post-Training