GroupChat handle is_termination_msg

[afourney]

Why are these changes needed?

Setting the is_termination_msg function handler in the GroupChatManager has no effect. This is because the GroupChatManager simply forwards the message to the next agent, and it falls on those agents to stop the conversation. This is problematic as it wastes orchestration LLM calls when the conversation is over, and because it prevents the ability to use the GroupChatManager as a central place to control group chat termination.

This PR fixes the issue by checking the termination message before rebroadcasting.

Related issue number

Closes #802

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. See https://microsoft.github.io/autogen/docs/Contribute#documentation to build and test documentation locally.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

[codecov-commenter]

Codecov Report

Attention: Patch coverage is 50.00000% with 2 lines in your changes missing coverage. Please review.

Project coverage is 46.29%. Comparing base (f96963e) to head (1272a69).
Report is 1574 commits behind head on main.

Files with missing lines	Patch %	Lines
autogen/agentchat/groupchat.py	50.00%	2 Missing ⚠️

Additional details and impacted files

@@             Coverage Diff             @@
##             main     #804       +/-   ##
===========================================
+ Coverage   27.76%   46.29%   +18.52%     
===========================================
  Files          27       27               
  Lines        3493     3497        +4     
  Branches      791      835       +44     
===========================================
+ Hits          970     1619      +649     
+ Misses       2452     1713      -739     
- Partials       71      165       +94     

Flag	Coverage Δ
unittests	46.06% <50.00%> (+18.35%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[victordibia]

Thanks for adding this @afourney .
The justifications here makes sense

• enabling better conversation control via the GroupChatManager
• better mental model on GroupChat behavior .. its easier to reason about setting termination logic via the central groupchatmanager than a specific agent or all agents
• also improves efficiency when the groupchatmanger can terminate conversations early/ correctly.

@afourney Any chance may be side effects here e.g. GroupChat notebooks where some results might benefit from multiple back and forths?

Approving in the mean time.

[afourney]

There should not be any side effects. The superclass default for is_termination_msg is None. In that case, this new PR acts identically to the published version. It's only when setting is_termination_msg to something other than None that a change in behavior occurs. If other notebooks were setting is_termination_msg for the GroupChatManager then this is actually a problem for those notebooks, because they were propagating the false belief that the is_terminiation_msg would be considered. In those cases, the revised behavior will indeed differ (and will likely more closely align to the author's intent). However, I'd argue than any such notebooks are already broken without this PR.