ENH: Add TrainerConfig to train/learncurve/eval/predict

[NickleDave]

This is part of the fix for #687. I think we should just be explicit we're using Lightning

• Change 'device' option to 'accelarator' in config files
• Rename 'get_default_device' -> 'get_default_accelerator'
• Define 'accelerator' in docs, link to Lightning docs

[NickleDave]

Renaming this to reflect that we want a table representing lightning.Trainer arguments in the config.

This will help address #742 and #745 as well.

We define a sub-table trainer for the train/learncurve/eval/predict configs

accelerator = 'cpu'

It should default to accelerator=get_default() and devices=1 to fix the above issues, and for now we will need to raise informative errors if someone tries to run on multiple devices.