When auxiliary.title_generation uses a reasoning model (MiniMax-M2.7, DeepSeek-R1, Qwen-QwQ, etc.), the auto-generated session title is empty because the code reads response.choices[0].message.content directly — which is None when the model returns content via structured reasoning fields instead of .content.

Fix: Replace the direct .message.content access with the existing extract_content_or_reasoning() helper that already handles reasoning models correctly. This is the same pattern used in session_search_tool.py, web_tools.py, and vision_tools.py.

Changes:

• Add extract_content_or_reasoning to the import from agent.auxiliary_client
• Replace response.choices[0].message.content with extract_content_or_reasoning(response)

Verification:

• py_compile passes
• All 19 existing title generator tests pass

Closes #18529