Responses API: Add presence_penalty parameter support

[leseb]

🚀 Describe the new functionality needed

Add support for the presence_penalty parameter in the /responses API endpoint (both request and response).

This parameter penalizes new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.

OpenAPI property paths:

• POST.requestBody.content.application/json.properties.presence_penalty
• POST.responses.200.content.application/json.properties.presence_penalty

💡 Why is this needed? What if we don't build it?

OpenAI API conformance is crucial for Llama Stack adoption. The presence_penalty parameter is commonly used to encourage topic diversity.

Without this parameter:

• Users cannot control topic diversity when using the Responses API
• Applications migrating from OpenAI may require code changes
• The Responses API conformance score (currently 80.9%) will remain lower than it could be

Other thoughts

Reference: OpenResponses OpenAPI Spec

[nathan-weinberg]

I can take this one as well