Add context as which specific api formats (openai-responses/completions) a model supports #769
Description
Mostly focused on OpenAI responses vs completions here, but can imagine some other major provider could change the format in the future.
Old (gpt-4) models only support completions. Some new-ish models support both responses and completions (gpt-5). Some support only responses (5.2-codex). Would be great to have context as to which specific format(s) a model supports.
Maybe api_formats: ["openai-responses", "openai-completions"]