Incorrect MisconfigurationException for models without dataloaders.

[yukw777]

🐛 Bug

I have a model that does not have train, val and test dataloaders defined internally (it's a production system and it doesn't really make sense to have dataloaders). If I try to run fit() on it by passing in train_dataloader and val_dataloaders, it raises

pytorch_lightning.utilities.exceptions.MisconfigurationException: You have defined `test_step()`, but have not passed in a `test_dataloader()`.

This means that it's now impossible to train a model without dataloaders defined, as there's no way of passing in test dataloaders. I believe this was caused by this PR: #1434. This is happening at the tip of master.

To Reproduce

Steps to reproduce the behavior:

1. Checkout the master branch
2. Define a model without data loaders
3. Run fit() with train_dataloader and val_dataloaders
4. See the exception

Code sample

# MyModel doesn't have train_dataloader, val_dataloader or test_dataloader
model = MyModel()
trainer = pl.Trainer()
trainer.fit(model, train_dataloader=train, val_dataloaders=val)  # exception raised here
trainer.test(test_dataloaders=test)

Expected behavior

There should be no exception raised during fit().

Environment

cuda:
	GPU:
	available:           False
	version:             None
packages:
	numpy:               1.18.1
	pyTorch_debug:       False
	pyTorch_version:     1.4.0
	pytorch-lightning:   0.7.4-dev
	tensorboard:         2.1.1
	tqdm:                4.43.0
system:
	OS:                  Darwin
	architecture:
		64bit
		
	processor:           i386
	python:              3.7.3
	version:             Darwin Kernel Version 18.7.0: Mon Feb 10 21:08:45 PST 2020; root:xnu-4903.278.28~1/RELEASE_X86_64

Additional context

[yukw777]

Oops, it seems like it's already being fixed! #1495, thanks @williamFalcon!!