Add AG2 multi-agent example with Mistral models

[faridun-ag2]

Summary

• Adds an AG2 (formerly AutoGen) integration example in third_party/AG2/
• Demonstrates multi-agent orchestration with Mistral models via three progressive examples
• Updates root README third-party tools table with AG2 entry

What's included

third_party/AG2/ag2_multi_agent_mistral.ipynb — Colab-compatible notebook with:

1. Basic two-agent conversation — AssistantAgent + UserProxyAgent with Mistral
2. Function calling — AG2's dual registration pattern (register_for_llm + register_for_execution)
3. Multi-agent GroupChat — automatic speaker selection with specialist agents

How it differs from MS_Autogen_pgsql

The existing third_party/MS_Autogen_pgsql/ example uses Microsoft AutoGen (the original project).
AG2 is the community-maintained fork with:

• Active development (500K+ monthly PyPI downloads)
• Native Mistral support via MistralLLMConfigEntry
• GroupChat multi-agent orchestration
• pip install ag2 (separate package from Microsoft's autogen-agentchat)

Test Plan

• Notebook runs in Google Colab
• pre-commit run detect-private-key --all-files passes
• No hardcoded API keys
• All Mistral API calls use os.getenv("MISTRAL_API_KEY")
• Pinned package versions
• Root README table updated
• Existing examples untouched

[faridun-ag2]

Hey @jaccolor2

This PR adds an AG2 integration example to the cookbook. Here's a quick overview of what's included:

Notebook: third_party/AG2/ag2_multi_agent_mistral.ipynb

Three progressive examples demonstrating AG2 + Mistral:

1. Basic two-agent conversation — Shows the fundamental AssistantAgent + UserProxyAgent pattern with Mistral via MistralLLMConfigEntry. Minimal setup, instant results.

2. Function calling — Demonstrates AG2's dual registration pattern: register_for_llm() tells Mistral about the tool schema, register_for_execution() tells the agent how to run it. Uses simulated weather/time tools to keep it self-contained.

3. Multi-agent GroupChat — AG2's flagship feature: multiple specialist agents (Researcher, Writer, Critic) collaborate in a group chat where Mistral automatically selects the next speaker. No manual routing needed.

Key details:

• Colab-compatible with pinned dependencies (ag2[openai]==0.11.4, mistralai==2.1.3)
• No hardcoded keys — all via os.getenv("MISTRAL_API_KEY")
• Uses AG2's current API: MistralLLMConfigEntry for config, .run().process() for execution
• Root README table updated with the AG2 entry
• No existing files modified beyond the README table

AG2 (formerly AutoGen) is a separate project from the existing MS_Autogen_pgsql example — it's the community-maintained fork with native Mistral support and 500K+ monthly PyPI downloads.

Happy to adjust anything as needed!