Implement the parquet output step

Parquet has compression and indexing while being logically not that different to a CSV from the user's point of view. We should output the CSV data to it in both original and pseudonymised version.

Conversion from Arrow is very easy apparently, if we go that route (see #15 )