Goal: 

Currently, the post_training API, specifically supervised_fine_tune requires arguments that aren't always used in out of tree providers such as `DataConfig`, `AlgorithmConfig`, or even `Model`. These are fields that are validated in strict ways inside of LLS and should be marked as optional if the provider implementation itself aims to handle these.

Acceptance Criteria:

Fields are audited in the supervised_fine_tune method and the `TrainingConfig` class in general. After auditing, specific fields should be marked as optional upstream