Add a :next-agent to the agent config

[bhauman]

The idea here is that if an agent has the :next-agent key defined we take the output of this agent and pipe it into the agent specified by this next-agent key.

This will just be a loop that follows these keys until there is no longer a next-agent key. At this point it returns the output of the last agent or perhaps it returns the output of all the agents or we stash agent outputs in the scratch pad under named keys and allow the parent model to decide what it wants to look at. In this last case we return the last result with a list of keys to the intermediate data.

Or we could rely on the model to store intermediate data in the scratch pad for us.

I'm thinking the simplest is to just return the last agent output with the full results stored under a specific reusable key in the scratch pad for the parent LLM to inspect if needed.