make ConstrainedOutputModel compatible with nnvision.utility.measures:model_predictions

[MaxFBurg]

Before, calling model_predicitons to get the predictions of a ConstrainedOutputModel resulted in a bug, as data_key was in self.forward_kwargs but also in kwargs

[MaxFBurg]

Code example to reproduce the issue:

        dataloaders, model = self.model_loader.load(key=key)
        output_selected_model = self.selector_table().get_output_selected_model(model, key)
        data_key = key['data_key']
        dataloader = dataloaders["train"][data_key]
        predictions = model_predictions(output_selected_model, dataloader, data_key, device=self.device)[1]

[MaxFBurg]

Hi @KonstantinWilleke @PPierzc I just ran into this problem again, realizing that this PR was never reviewed. Could you please review and merge it?

[MaxFBurg]

... this one would actually be important to be merged soon, as my follow up PR's might lead to merge conflicts that we could avoid