The `prefer` parameter: routing AI requests by intent instead of model name

Here's a pattern I see everywhere in LLM code: response = openai.chat.completions.create( ...