Disentangle handling of preprocessing

[yannikschaelte]

Feature description
Some applications (e.g. learning sumstats) require a more involved analysis in preprocessing, which cannot be fully automatized. Thus, it should be possible to

• have a separate function for preprocessing
• store the preprocessing samples for re-use in the database
• allow it to be of a different size than the population size
• still allow to automatically run preprocessing in the abcsmc run at the beginning