Split from !1255 (closed)
An interesting side-effect of this change is that subjects who do not have an attributeset can still be recruited (if no subjectfilters are set for the study). We were never really transparent about the (non-)existance of an attributeset, so I think this is fine -- maybe even less confusing.