🚀 Describe the new functionality needed

Logprobs are an important tool to help users understand token prediction outcomes. They are currently missing from llama stack responses.

Requirements

• The Responses Create requests has a top_logprobs field, which gets translated or dropped based on the the inference provider.
• The response object contains top_lobprobs which is equal to the output value from the provider, or None.

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

Users that have max_logprobs set will get bad request errors.

Other thoughts

No response