[Feature Request] Add streaming support for structured generation

[rehan243]

Feature Description

Enable streaming output while maintaining JSON schema enforcement, yielding valid partial JSON as it's generated.

Motivation

For UX-sensitive applications (chatbots, real-time dashboards), streaming responses while maintaining structure would improve perceived latency.

Current Limitation

Structured generation currently requires waiting for the full output before validation/delivery.

Proposed Approach

Stream tokens while maintaining a running FSM state, yielding complete sub-objects as they become valid.

This would combine the best of structured output guarantees with the UX benefits of streaming. Great project!

[RobinPicard]

Hi @rehan243! That's something we've been thinking of doing, thanks for opening an issue about it!