New Integration: AG2 multi-agent framework

Summary

Closes #3107

Add an AG2Agent Haystack component that wraps AG2 (formerly AutoGen) multi-agent conversations, allowing Haystack pipelines to leverage AG2's multi-agent orchestration.

What's included

• AG2Agent — a Haystack @component wrapping AG2 multi-agent conversations
• Full serialization support (to_dict / from_dict)
• Configurable: model, system message, API type, code execution, auto-reply limits
• Unit tests with mocked dependencies
• README with standalone and pipeline usage examples

How it works

On each run(), the component creates an AG2 AssistantAgent and UserProxyAgent, executes the conversation via user_proxy.run(assistant, message=query).process(), and returns the final reply along with the full message history.

from haystack import Pipeline
from haystack_integrations.components.agents.ag2 import AG2Agent

pipeline = Pipeline()
pipeline.add_component("agent", AG2Agent(model="gpt-4o-mini"))
result = pipeline.run({"agent": {"query": "Explain RAG."}})
print(result["agent"]["reply"])

Start at the AG2Agent entry point in haystack_integrations.components.agents.ag2 and review the issue's standalone and pipeline usage examples. Implement the component with the listed configuration, serialization, conversation, reply, and message-history behavior, then verify it with mocked unit tests and confirm the README examples work.