The relevant data is copied into memory in pretty large chunks, so the core driver is not expected
to do much for us here.  In some brief testing, using the default driver instead resulted in
approximately the same (or slightly better) runtime and significantly lower peak memory consumption.

This requires the pending MR for cal_db_interactive.  Tested with e79f2c9d (current head of the
branch of that MR).

Applying isort and nbstripout.  Changing from ipyparallel to multiprocessing - and while doing so,
simplifying the call signature of characterize_module.

This metadata file subsumes the old retrieved constants YAML file.  The commit includes some refactoring centered around pathlib.