Remove logging of figures

[robmarkcole]

Per the comment in #2138 remove figure logging

[adamjstewart]

Hmm, not sure how to test datamodule plotting now that this is gone. Not even sure if we need datamodule plotting anymore.

[adamjstewart]

Some datamodules also have custom plot methods.

[adamjstewart]

Did we decide we want to remove datamodule plotting, or just find a different way to test it? Not sure if @isaaccorley's callback suggestion still requires datamodule plotting methods.

[isaaccorley]

I would be in favor of just keeping the datamodule.plot as a reference to the dataset so users don't have to dig to find out how to access the dataset to plot. But not actually testing it if it's just a reference to the dataset plot method.

Then it would be clear that a user can do trainer.datamodule.plot without having to do trainer.datamodule.<split>_dataset.plot

[adamjstewart]

Well we have to test it if we want to maintain 100% coverage. Can you share an example callback you use?

[adamjstewart]

Ping @isaaccorley, we need a callback to get test coverage if you want to keep datamodule.plot() methods in place.

[isaaccorley]

100% approve of this

[adamjstewart]

We can't merge without full coverage. We can't get full coverage without testing the plotting or removing the plotting.

[robmarkcole]

@adamjstewart is testing or removing plotting in scope of this PR? Seems a shame to remove it, but also not sure how you would even test plotting

[adamjstewart]

I can't merge without 100% test coverage, which makes it within the scope of this PR. Testing would involve adding a callback to our trainer tests to do what our trainers used to do manually.