responses: expose image generation output metadata

[jeet-dekivadia]

Summary

• expose action, background, output_format, quality, revised_prompt, and size on image-generation response variants
• keep response input-item, item, output-item, and conversation-item representations consistent
• add a decode regression through ResponseOutputItemUnion.AsImageGenerationCall()

The Responses API image-generation guide documents these fields in emitted image_generation_call output items: https://developers.openai.com/api/docs/guides/tools-image-generation

Fixes #675.

Validation

• git diff --check
• go test ./responses -run '^TestResponseOutputItemImageGenerationCallFields$' -count=1

Remote Linux validation

• git diff --check
• ./scripts/lint
• ./scripts/bootstrap
• ./scripts/test

[jeet-dekivadia]

Small follow-up while this is waiting for review. I kept the generated-shape change focused on image-generation response metadata that is already present on the API object, without changing request serialization. The added regression checks the concrete image-generation output fields so callers can read revised_prompt, size, quality, background, output_format, and action instead of losing that metadata.