Significant speed improvements to `preprocess`

[mir-am]

The speed of preprocess is quite significantly improved by using parallel_apply() for processing functions' arguments and reducing the depth of parametric types. parallel_apply() is used from pandarallel.