Add RLOO and REINFORCE++ Advantage Estimators as GRPO Extensions

[ved1beta]

⚠️ Please check that this feature request hasn't been suggested before.

• I searched previous Ideas in Discussions didn't find any similar feature requests.
• I searched previous Issues didn't find any similar feature requests.

🔖 Feature description

RLOO (REINFORCE Leave-One-Out) and REINFORCE++ are closely related online-RL algorithms that share the same rollout-and-reward structure but use different baselines/advantage estimation

✔️ Solution

-Add an advantage-estimator selector either by extending loss_type

• Implement the differing advantage computation in the GRPO strategy/trainer

❓ Alternatives

NA

📝 Additional Context

NA

Acknowledgements

• My issue title is concise, descriptive, and in title casing.
• I have searched the existing issues to make sure this feature has not been requested yet.
• I have provided enough information for the maintainers to understand and evaluate this request.