68 bug welldatamodule normalization not applied to val and test data

[mikemccabe210]

Fixes pattern where datamodule inconsistently normalizes data and valid/test normalization is done in trainer.

Changes:

• Pass normalization to val/test and rollout versions
• Change normalization functions in trainer to have consistent API (optional batch dict, tensor inputs returning both as output)
• Now explicitly denormalizes target in valid/test rather than expecting non-normalized targets as input.
• Increment hotfix version number

[mikemccabe210]

@kazewong I'm assuming you're out for the day, but please sanity check this whenever you get to it. Still running some checks to make sure I didn't break anything so i'm not going to merge the second you approve.