Restoring local model does not load mean and std

Description:
When specifying a local path to a pre-trained .pth model via the restore_model field in experiments.csv, the expected mean and std values are not loaded. As a result, the training process fails with:

Exception: No file found at f{mean_path}

This happens because the logic in experiment.py only calls download_model() (which sets and saves the normalization parameters) if restore_model does not end with .pth: https://gitlab.teklia.com/dla/doc-ufcn/-/blob/main/doc_ufcn/train/experiment.py?ref_type=heads#L253
Therefore, if a user specifies a local .pth model, the mean and std files are not generated, but are still expected later in get_mean_std().

Steps to Reproduce:

• Set restore_model in experiments.csv to a local path
• Remove any existing mean and std files
• Run the training script
• Observe the error in get_mean_std() due to missing files

Are there still things that I have not taken into account or is loading the values from an existing model incorrect?

Thank you very much in advance!

Best regards,
Constantin