Add Multi-Agent workflow with split manager/developer responsibilities

[aaronsteers]

While attempting to make a fully autonomous agent, I experienced continued issues with the agent saying "I'm done" before it was actually done. To better handle this workflow, the new model is:

1. A manager agent responsible for quality control and deciding when we're "done" or "blocked/failed".
2. A developer agent that simply keeps working on the problem, with instructions/guidance from the manager.

Tools distribution in this model:

1. Only the manager has access to call the job successful or failed (blocked).
   • In order to ensure this is explicit, a special pair of tools exist specifically for the manager to declare success or failure.
2. Both the manager and developer can access the session's project files, and both can run the connector_readiness_check tool.
3. Only the developer has access to low-level operations like manifest validation and running test reads.

Under consideration:

• We could create a dedicated agent for research, responsible for getting docs, and then would be the only agent with direct internet access. (Better security model.)