Save intermediate query results and durations for accuracy plot

[mIXs222]

Currently run_query only saves the last df query result. We would want intermediate results to measure accuracy over partition/time. These changes are

1. Save possibly multiple intermediate dfs into multiple CSVs
2. Save durations for these results into JSON (for parsing/inspecting convenience)

Add some error handling too, e.g. the panic message is more clear when outputs directory doesn't exist.