[Python] SelectorGroupChat to support concurrent speakers

[ekzhu]

SelectorGroupChat can support concurrent speakers by choosing multiple speakers at once, and have the manager waiting for all the selected speakers to respond before selecting the next round of speakers.

Related discussion: #5364

[CrankGentleman]

@ekzhu I also hope to execute multiple agents simultaneously. In my own project, there are agents for weather, air tickets, and hotels. However, I need it to initiate the parallel execution of three agents and then generate output. Clearly, this feature is currently not supported. This functionality is crucial as it can significantly shorten the waiting time for the entire application.

[ekzhu]

It is actually not hard to add the support:

1. Make select speaker method return multiple speakers as a list: update the selector_func signature to return str | List[str] | None.
2. Update the default selection function to handle multiple speakers when the model returned more than one speaker names.
3. In SelectorGroupChatManager, rather than publishing only one RequestToSpeak event, publish one for each selected agent's topic type.
4. In SelectorGroupChatManager, buffer agent responses when there are multiple concurrent agent speakers, only start checking termination conditions etc after all speaker responded.

3 and 4 requires some refactoring of BaseGroupChatManager and have SelectorGroupChatManager override some methods in the base.