Built-in AI inference

Provide first-party AI inference primitives so teams can run prompts, structured generation, and model workflows inside Convex without assembling separate infrastructure.