Rows are currently stored on disk in this format:
You have to consult the accompanying metadata JSON file to find out what the column names are.
A command that streams out one of these dumps as newline separated objects with the column names as keys would be useful.