fix multi_GPU training, support pytorch2.0

[KdaiP]

• Some parameters have already written in hps.model (vits2_vctk_base.json), these would cause error TypeError: type object got multiple values for keyword argument 'use_transformer_flows'
• Set return_complex to False in order to enable training in PyTorch 2.0. This change is necessary due to PyTorch 2.0 modifying the default value in stft from False to True.
• Add a tqdm progress bar