Python: Rejected system messages leave AgentChat permanently active #14638

AgentChat.add_chat_messages() sets its activity signal before validating message roles. If it rejects a system message, it raises before entering the try/finally that clears the signal. After the caller catches this validation error, the same AgentGroupChat cannot accept a corrected message, read its history, or reset: each operation reports that another agent is active.

Reproduced against current main, commit 25501830b9483ad6b23955dee340c8590fa54152, with Python 3.12.13 on macOS arm64:

import asyncio

from semantic_kernel.agents import AgentGroupChat
from semantic_kernel.contents import ChatMessageContent
from semantic_kernel.contents.utils.author_role import AuthorRole
from semantic_kernel.exceptions.agent_exceptions import AgentChatException


async def main():
    chat = AgentGroupChat()
    await chat.add_chat_message("seed")

    try:
        await chat.add_chat_message(
            ChatMessageContent(role=AuthorRole.SYSTEM, content="unsupported")
        )
    except AgentChatException:
        pass

    print(chat.is_active)  # True, although no operation is running
    await chat.add_chat_message("corrected input")
    # Exception: Unable to proceed while another agent is active.


asyncio.run(main())

Expected behavior: keep rejecting system messages, preserve the existing history, and release the activity signal so the caller can recover. A mixed user/system batch is correctly rejected without partially appending the user message, but it leaves the same stuck activity signal.

The existing test_add_chat_messages_throws_exception_for_system_role checks the rejection but does not check recovery. The existing group-chat and strategy tests pass (77 tests) while six real-API rejection/recovery scenarios reproduce the problem and three valid-message controls succeed. These checks require no model or provider API.

I would like to take on a small fix that keeps role validation inside the activity cleanup scope, with regression coverage for history preservation and recovery. The activity acquisition would remain outside the try so a rejected competing operation cannot release another operation's signal. Is this correction still welcome in the legacy group-chat API given the Agent Framework migration?

This issue and the offline reproduction were prepared with Codex assistance. No implementation patch has been submitted.