ai ai@7.0.168

Version 7.0.168 of `ai`, the core package of Vercel's AI SDK, adds an answer to a common chat-interface question: what happens when a user sends another message while the assistant is still answering. `useChat` is the SDK's hook for building chat interfaces; it keeps the message list and streams responses into it. Per the release notes, its new `experimental_queue` option queues messages sent while a response is in progress. The `experimental_` prefix marks an API that may still change, and the next release, ai@7.0.169, already fixed how the queue behaves around tool approvals.

The other change fixes `toUIMessageStream`, which converts a model's output into the message stream a chat interface reads. Its `onError` callback is now called once for each invalid tool call. The fix's wording indicates that it could previously fire more than once for the same call, so applications that log or count errors in that callback may see fewer duplicate reports after upgrading. The remaining entries update `@ai-sdk/gateway` to 4.0.141 and `@ai-sdk/provider-utils` to 5.0.78.