add return dict false test and bug fix

[dakinggg]

What does this PR do?

Fixes an issue introduced in 0.12.1 where HuggingFaceModel crashes if config.return_dict = False. The code assumed that that outputs was None in eval_forward, but this is not the case during training, when the cached train outputs are used. Also adds a test for this setting.

What issue(s) does this change relate to?

Closes CO-1743

Before submitting

• Have you read the contributor guidelines?
• Was this change discussed/approved in a GitHub issue first? It is much more likely to be merged if so.
• Did you update any related docs and document your change?
• Did you update any related tests and add any new tests related to your change? (see testing)
• Did you run the tests locally to make sure they pass?
• Did you run pre-commit on your change? (see the pre-commit section of prerequisites)

[alextrott16]

LGTM