[Discussion] support multi-agent systems.

[kerthcet]

What would you like to be added:

Gradually become a multi-agent observability system, for example, we should have a agent registry. The difficult part would be how to design the system, if we want to be a multi-agent observing platform, seems it has very little relationship with the experiments because agents are long-running systems, experiments are usually go-to-end tasks. Does that mean we will have two different kinds of workloads, one is experiment, another is agent services.

However, for hive, sandboxes are also agents, they're also running experiments.

Why is this needed:

Completion requirements:

This feature requires the following artifacts:

• Design doc
• API change
• Docs update

The artifacts should be linked in subsequent comments.