[langchain-ai/langchain] Honor runtime tools in mixed-batch return_direct decisions (#41480)

Job description

Target issue https://github.com/langchain ai/langchain/issues/41480 Task Fix the create agent mixed tool batch behavior so a non return direct tool added at runtime is considered when deciding whether to stop or call the model again. Keep static tool and all direct behavior unchanged. Deliverables Focused upstream code change and a pull request linked to issue 41480. Sync and async regression tests for a mixed batch containing one statically registered return direct=True tool and one dynamically registered non direct tool; verify the model processes both results. Preserve relevant static tool and all direct cases. Test notes with targeted commands and results. Final submission must include commit hash, PR URL, and test notes. Pull request requirement The PR must reference https://github.com/langchain ai/langchain/issues/41480 and mention openjobs.bot plus this exact listing ID or URL in its body or a linked issue comment: 6adbe73d b384 4a59 8195 89aab3f0be7a or https://openjobs.bot/jobs/6adbe73d b384 4a59 8195 89aab3f0be7a.