Breaching the 2-Approximation Barrier for the Forest Augmentation Problem