fix: preserve single-turn structured output

[he-yufeng]

Fixes #6089.

Summary

• Keep message_as_output for plain single-turn agent text output.
• Leave structured output_schema results as event.output instead of marking them as message output.
• Add direct tests for both plain text and structured single-turn output handling.

To verify

• $env:PYTHONPATH=(Resolve-Path .\src).Path; python -m pytest tests\unittests\workflow\test_llm_agent_as_node.py -q
• python -m py_compile src\google\adk\workflow\_llm_agent_wrapper.py tests\unittests\workflow\test_llm_agent_as_node.py
• python -m ruff check src\google\adk\workflow\_llm_agent_wrapper.py
• git diff --check

[rohityan]

Hi @he-yufeng , Thank you for your contribution! We appreciate you taking the time to submit this pull request. Can you please fix the failing unit tests before we can proceed with the review.

[he-yufeng]

@rohityan Done, and sorry for the long silence on your June ask. Rebased onto current main (the test file got a union merge with the branch-coverage tests landed since) and the workflow unit suite is green locally: 735 passed, 11 skipped. The one thing I could not replay is the original CI failure, since the June run logs are gone; if a specific test still fails on your side, name it and I'll fix that one directly.

[he-yufeng]

Still applies on main: process_llm_agent_output sets message_as_output=True unconditionally, overriding what the caller asked for. Could a maintainer re-review when there's a chance?

[adk-bot]

Thank you @he-yufeng for your contribution! 🎉

Your changes have been successfully imported and merged via Copybara in commit 6b96ba1.

Closing this PR as the changes are now in the main branch.