Trainers: support nn.Module as input

[adamjstewart]

Want to use a model that does not yet exist in TorchGeo? Now you can pass it directly to the trainer class.

• Code
• Testing

[adamjstewart]

Unit tests have revealed that this idea doesn't actually work. We can't store model in self.model because we haven't yet called super().__init__(), but that calls configure_models which needs access to self.models. And we can't store model in self.hparams because it doesn't support nn.Module. Need to think of a clever way to break this dependency loop...