Add AG2 multi-agent notebooks to third_party examples

[faridun-ag2]

Summary

• Adds two notebooks demonstrating AG2 multi-agent capabilities with OpenAI models to examples/third_party/
• multi_agent_chat_with_ag2.ipynb — Two-agent conversation and multi-agent GroupChat with automatic speaker selection (Researcher → Writer → Critic collaboration via GPT-4o-mini)
• tool_use_with_ag2_agents.ipynb — Agent tool use via AG2's decorator pattern, mapping directly to OpenAI function calling (stock lookup, compound interest calculator)

AG2 is an open-source multi-agent framework with 500K+ monthly PyPI downloads, 4,300+ GitHub stars, and 400+ contributors. It uses OpenAI models as the default LLM backend.

Why this fits the cookbook

• OpenAI models power the agents — demonstrates production patterns for function calling + multi-agent orchestration
• Follows existing third_party/ conventions (cf. LlamaIndex, W&B notebooks)
• Progressive complexity: simple chat → GroupChat → tool use

Test plan

• Both notebooks execute top-to-bottom without errors (jupyter execute)
• Two-agent chat produces clean response with TERMINATE
• GroupChat shows multiple agents taking turns (Researcher → Writer → Critic)
• Tool use notebook shows function calls being made and correct results returned
• No hardcoded API keys (os.environ["OPENAI_API_KEY"])
• Execution counts cleared before commit
• registry.yaml and authors.yaml updated
• Passes notebook validation check (check_notebooks.py)
• Follows PEP 8, four-space indentation

[faridun-ag2]

Hi @ted-at-openai @colin-openai !
I would love a review on this when you get a chance. These notebooks demonstrate AG2 multi-agent workflows powered by OpenAI models, following the existing third_party/ conventions.
Happy to address any feedback!