Fix incorrect handling of on_batch_end edge cases in run_training_batch

[jeffling]

Before submitting

• Was this discussed/approved via a Github issue? (no need for typos, doc improvements)
• Did you read the contributor guideline?
• Did you make sure to update the docs?
• Did you write any new necessary tests?

What does this PR do?

This fixes a bug

ValueError: not enough values to unpack (expected 3, got 2)

When trying to return -1 to exit early during on_batch_start, and also when the batch is empty.

For future work, we should have a test for this regression.

https://github.com/williamFalcon/pytorch-lightning/blob/8f8cea1c5759524c6fc2eb33b972bba64e5ce0f4/pytorch_lightning/trainer/train_loop_mixin.py#L100

PR review

Anyone in the community is free to review the PR once the tests have passed.
If we didn't discuss your PR in Github issues there's a high chance it will not be merged.

Did you have fun?

Make sure you had fun coding 🙃

[jeffling]

It's also very unfortunate that we initialize all_log_metrics as a list and turn it into a dict somewhere. Ideally, things more or less stay the same type. However, I don't have time to clean it up now, just noting this.

[Borda]

I would even change its call from

output = self.run_training_batch(batch, batch_nb)
batch_result, grad_norm_dic, batch_step_metrics = output

to

batch_result, grad_norm_dic, batch_step_metrics = \
    self.run_training_batch(batch, batch_nb)

[jeffling]

I think we can do as as a followup, I'd like to fix the nit I pointed as well. Right now this is breaking documented functionality so we should merge this in as a hotfix and do a neatness refactor later

[williamFalcon]

@jeffling thanks! please follow up on the issue you mentioned :)