[QoIAwareProjection] Fix discrepancy between the current basis and the training and validation loss computed using that basis
[QoIAwareProjection
] Fix discrepancy between the current basis and the training and validation loss computed using that basis. The current inconsistency stems from the fact that the training loss is computed before applying optimization at each epoch and the validation loss is computed after applying optimization. The best basis is now shifted by one index compared to the minimum training loss.