Self-Correcting Self-Consuming Loops for Generative Model Training