feat(ag2): upgrade to v0.11.4 and rewrite all examples with 7 new additions

[martimfasantos]

Summary

• Upgrade AG2 from 0.9.1.post0 to 0.11.4 — rewrite all 5 existing examples for the new API and add 7 new examples, bringing the total to 12 hands-on examples
• All examples follow repo conventions (zero-padded filenames, dashed-border docstrings, section headers, settings-based config)
• Includes verified OUTPUTS.md with captured output for all 12 examples

Examples

#	File	Feature
00	00_simple_agent.py	Basic ConversableAgent, run() / process(), LLMConfig
01	01_agent_with_tools.py	functions= parameter, tool registration, multi-tool agents
02	02_structured_outputs.py	Pydantic response_format, schema-conforming JSON responses
03	03_human_in_the_loop.py	Simulated human approval, register_reply(), conversation flow
04	04_multi_agent.py	AutoPattern, initiate_group_chat(), multi-agent collaboration
05	05_sequential_chat.py	initiate_chats() pipeline, carryover summaries, specialist agents
06	06_nested_chat.py	register_nested_chats(), encapsulated workflows, inner delegation
07	07_code_execution.py	LocalCommandLineCodeExecutor, code writer/executor agent pattern
08	08_guardrails.py	Safeguard policies (Maris), regex filtering, inter-agent safeguards
09	09_mcp_tools.py	create_toolkit(), stdio_client, FastMCP server integration
10	10_observability.py	runtime_logging to SQLite, event capture, post-hoc analysis
11	11_a2a.py	A2aAgentServer, A2aRemoteAgent, distributed agents over HTTP

Key API changes (0.9.x → 0.11.4)

• LLMConfig now takes a positional dict: LLMConfig({"model": name}) instead of keyword args
• New agent.run() / result.process() pattern replaces direct initiate_chat() for simple cases
• functions= parameter on ConversableAgent for inline tool registration
• Group chat uses AutoPattern + initiate_group_chat() (old GroupChat/GroupChatManager still supported)
• Guardrails (Maris) via apply_safeguard_policy() with regex-based inter-agent safeguards
• A2A protocol support with A2aAgentServer and A2aRemoteAgent
• MCP integration requires async a_initiate_chat() to avoid blocking the stdio transport event loop

Configuration

• All examples use OpenAI (gpt-4o-mini) via settings.py
• Dependencies: ag2[openai,mcp,a2a]>=0.11.4, mcp>=1.9.2, pydantic-settings, uvicorn
• Managed with uv package manager

Testing

All 12 examples ran successfully with exit code 0 and produced feature-showcasing output (see OUTPUTS.md for full details).