Implement a Supervisor-Agent Workflow

Implement run_supervisor(goal, workers, tasks, max_steps) that manages a multi-worker task execution workflow. workers is a dict mapping worker names to callable functions. tasks is a list of dicts, each with: - id (str): unique task identifier. - worker (str): name of the worker to dispatch to. - depends_on (list[str], optional): task IDs that must complete first. - params (any): arguments to pass to the worker. Requirements: - Each iteration, identify ready tasks (all dependencies in completed). - Dispatch each ready task to its assigned worker via workers[name](params). - Collect results in a completed dict by task ID. - On worker failure, mark the task as failed with None result but do not abort other tasks. - If a worker name is not in workers, mark the task as failed with an error message. - Stop when all tasks are done or max_steps steps have been used. - Return {"completed": dict, "trace": list, "steps_used": int, "all_completed": bool}. - The trace should contain one entry per task: {"task": str, "worker": str, "status": str, "result": any}.