Expose APIs from RL4LM to Consume 2 Models, 1 Scenario and 1 Morality Score to align the schema generator LM

[logisticloon]

The purpose of this subtask is to ensure Alignment of LM_SchemaGenerator

The two models are LM_SchemaGenerator and LM_Task.

The API should expect path of the two models (perhaps checkpoints and Another API callback function for LM_Task) and a scenario as input parameters.

Ensure a fork from RL4LM and not RL4F.

Subtasks:

• Extend the metric function to incorporate MSE ( Reuse your last implementation )
• PPO training setup for LM_SchemaGenerator.