Curate Data depending on agent

Beginning to think we should have a Data schema replacing describe_data and have different output context curated for each SQL, vega lite, etc agent, and maybe the agents decide how to dissect the data--if not just use default describe_data