Updated API with v0.2.0

Some restructuring to make Model more general. This means that some earlier code might break.

Main differences are:

• model.predict will use the model.net.predict method if available. If not, it will use the old model.net.forward.
• model.predict_net will use model.net.forward.
• Both predict methods now works with dataloader input.
• Removed predict_dataloader as predict works with dataloaders.
• Both predict methods take the optional argument func which can be a torch function that is applied to the output. E.g., func=torch.sigmoid can be used for binary classification.
• target is now an optional argument in fit.
• compute_metrics take the argument data instead of input, target.
• set_device can be used to change the device of the model. E.g., model.set_device('cpu') will enable the model to work on cpu.
• Renamed DataLoaderSlice to DataLoaderBatch.