[TODO]: use best model in notebooks

[KnathanM]

Our example notebooks use trainer.test(mpnn, test_loader). We may want to show users how to test on their best model. They would do this by making a checkpointing callback and passing that to the trainer:

from lightning.pytorch.callbacks import ModelCheckpoint
checkpointing = ModelCheckpoint(
           "tempdir",
            "best-{epoch}-{val_loss:.2f}",
            "val_loss",
            mode="min",
            save_last=True,
        )
trainer = pl.Trainer(callbacks=[checkpointing])

Then they don't give the trainer a model: trainer.test(dataloaders=test_loader).
We can do this as part of the documentation after v2.0 is out on pypi.