Text this: Enhancing neuromolecular imaging classification in low-data regimes with generative machine learning: A case study in HDAC PET/MR imaging of alcohol use disorder