[roadmap] Trinity-RFT development

[yanxi-chen]

• Support generic multi-step RL scenarios (infra & algorithm)
• Further optimize utilization of computational resources
• Support automatic load balancing
• Improve robustness of WorkflowRunner --> Upgraded to Scheduler
• Support training with Megatron-LM
• Implement more algorithms for off-policy / asynchronous RL
• Implement more advanced sampling strategies for task / experience buffer
• Further integrate RL process with advanced data processing functionalities
• Continue refining Trinity-Studio