Add wrapAISDKModel method for Vercel's AI SDK

[jacoblee93]

Improve support for tracing Vercel AI SDK calls with a new wrapAISDKModel method

Because AI SDK calls nest the text/object stream within the response, this PR adds an optional parameter to traceable that specifies a key in the response object to wrap. It also adds support for specially tapping ReadableStream so that traceable does not change them to plain async iterators.

Here's an example:

import { anthropic } from "@ai-sdk/anthropic";
import { streamObject } from "ai";
import { wrapAISDKModel } from "langsmith/wrappers/vercel";

const modelWithTracing = wrapAISDKModel(anthropic("claude-3-haiku-20240307"));
const { partialObjectStream } = await streamObject({
  model: modelWithTracing,
  prompt: "Write a vegetarian lasagna recipe for 4 people.",
  schema: z.object({
    ingredients: z.array(z.string()),
  }),
});
for await (const chunk of partialObjectStream) {
  console.log(chunk);
}

TODO: improve support in the UI for properly rendering token counts and aggregating chunks. The data is returned but is not in one of our currently supported formats.

Example traces:

Generate text: https://smith.langchain.com/public/fbd12847-9485-43cf-a0a3-82c0b3318594/r
Generate object: https://smith.langchain.com/public/c25944c7-2428-44d0-991b-ef3b5e47cab5/r
Stream text: https://smith.langchain.com/public/2fb34c85-fec5-4361-a487-ffcf67718750/r
Stream object: https://smith.langchain.com/public/27f5c9ba-006d-4a2f-b569-2e8da09dae23/r

CC @dqbd

[jacoblee93]

Will follow up with docs!