AI Gateway now supports asynchronous video generation

Vercel's asynchronous video support offers two different kinds of waiting. Adding polling or a webhook to generateVideo avoids holding one gateway request open throughout rendering, but the original calling process must still remain alive. The changelog distinguishes this from starting a job for later retrieval.

For work that must outlive the caller, startVideo returns an operation that the application can store in a database or queue. Another process can later pass it to getVideoStatus. That separates the rendering job from the lifetime of a serverless request. The status call returns hosted URLs or inline bytes; hosted videos need downloading before their URLs expire.