ETL is extract/transform/load. So basically you'd be taking your raw data, extracting it out of the files and transforming it into a sensible format or data structure, and loading it up into a persistent data store.

A million rows per column is much more reasonable than a million columns. That's still a ton of data depending on how many parent rows the associated rows have. I have some ideas but honestly it's probably best to get your co-workers feedback since they know the data and ask whatever other specific questions you have.