HF RL Explorer

DeepSeek Reasoner model fails because system message handling is too rigid

DeepSeek Reasoner model fails because system message handling is too rigid: a task in LegoFlow-SWE (Harbor dataset). A new utility class OpenAiModelUtils must be introduced to apply the chosen setting. Its public API should include:

The task

A new utility class **`OpenAiModelUtils`** must be introduced to apply the chosen setting. Its public API should include:

Part of Lego-X/LegoFlow-SWE.