Augmentation Invariant Discrete Representation for Generative Spoken Language Modeling