Support structured output for newer Anthropic models

[aoki-ryusei]

First of all, thank you for maintaining this awesome library!

I'm using the Anthropic provider with Active Agent.

Currently, structured output for the Anthropic provider appears to be implemented using the prompt engineering technique with prefilled responses.

However, starting with Claude Sonnet 4.5/4.6, prefilled responses are no longer supported, and requests using this approach now return a 400 error.

Reference:
https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices#migrating-away-from-prefilled-responses

Problem

As a result, structured output does not work with the latest Claude models.

Proposal

It would be great to support Anthropic's current structured output approach instead of relying on prefilled responses.

I've already prepared a PR with a proposed implementation. If this direction looks good, I'd really appreciate your feedback.

PR: #339

[TonsOfFun]

Thanks for putting this issue and the associated PR together, I'll review it today and let you know.

[TonsOfFun]

You can use this from activeagent main by updating your gem file accordingly. This will be included next time I cut a release. Thanks!

[aoki-ryusei]

Thanks so much for the quick turnaround!

I'll switch to main for now and test it out. Really appreciate you reviewing and merging this so quickly.

Looking forward to the next release. Thanks again!