Support filtering CKAN sources to avoid fetching irrelevant datasets

While import 100k datasets is nice to ensure that we scale at least a little bit, a lot of these datasets were not relevant for umwelt.info and hence I added a rudimentary filter functionality to restrict ourselves to environmental information. This should be especially useful when we merge the SNS AutoClassify integration as it will take a long take for the initial classification of all these datasets even when deduplicated.