Remove `*_step_end` from `LitModular`

[dxoigmn]

What does this PR do?

This PR merges *_step_end into *_step in LitModular. This means we no longer need to clear outputs.

This PR depends upon the following:

• [x] #169

Type of change

Please check all relevant options.

• [x] Improvement (non-breaking)
• [ ] Bug fix (non-breaking)
• [ ] New feature (non-breaking)
• [ ] Breaking change (fix or feature that would cause existing functionality to not work as expected)
• [ ] This change requires a documentation update

Testing

Please describe the tests that you ran to verify your changes. Consider listing any relevant details of your test configuration.

• [x] pytest
• [x] CUDA_VISIBLE_DEVICES=0 python -m mart experiment=CIFAR10_CNN_Adv trainer=gpu trainer.precision=16 reports 70% (21 sec/epoch).
• [x] CUDA_VISIBLE_DEVICES=0,1 python -m mart experiment=CIFAR10_CNN_Adv trainer=ddp trainer.precision=16 trainer.devices=2 model.optimizer.lr=0.2 trainer.max_steps=2925 datamodule.ims_per_batch=256 datamodule.world_size=2 reports 70% (14 sec/epoch).

Before submitting

• [x] The title is self-explanatory and the description concisely explains the PR
• [x] My PR does only one thing, instead of bundling different changes together
• [x] I list all the breaking changes introduced by this pull request
• [x] I have commented my code
• [ ] I have added tests that prove my fix is effective or that my feature works
• [x] New and existing unit tests pass locally with my changes
• [x] I have run pre-commit hooks with pre-commit run -a command without errors

Did you have fun?

Make sure you had fun coding 🙃

[dxoigmn]

I also introduced this change in #157 to work with newer PL.

Have you confirmed that there's no memory leak issue with the current PL version?

How did you confirm? I watched memory usage and it seemed fine. I also root caused (by reading internal PL code) why outputs.clear() was necessary in the old code. That is because PL is basically holding whatever training_step returns for training_epoch_end. This wasn't a problem for training_step (because we only returned a loss tensor) but was a problem for validation_step because we returned outputs there. Now we just return None and let metrics and log handle everything in their respective step functions (instead of using step_end).

[dxoigmn]

Note that RobustBench tests are failing because of some model issue unrelated to these changes...