Feature: handle large data sets via pointers

[philipp-baumann]

Once spectral libraries grow large and we do preprocessing, we might be bloating the random access memory. As an alternative, we could use a relational DB interface and a zarrr backend to store preprocessed data on disk.