[Proposal] Reasoning part for thinking models

[mbleigh]

There are several models now whose API responses includes their reasoning as part of the response.

I propose a new ReasoningPart:

type ReasoningPart = {reasoning: string};

that can be used to capture reasoning in message responses. So a response might look like:

{
  message: {
    role: "model",
    content: [
      {reasoning: "I am thinking about what I'm going to do..."},
      {text: "This is my actual answer."}
    ]
}