config sys for simple_grpo v1

[wwwjn]

Stack from ghstack (oldest at bottom):

• [rl] refactor torchtitan model registery in vllm #2194
• config sys for simple_grpo v2 #2193
• -> config sys for simple_grpo v1 #2192
• [rl] Use torchtitan config system for inference and simple GRPO #2191