Simple: because we receive and store every single message that is received and decoded. Yes, that's significantly more bandwidth (also for us, not to mention storage requirements) but that's the point of this whole undertaking as stated right on the front page. That raw data enables later large-scale research into many domains that require more than just (even fine-grained) trajectory data, i.e. "where does an aircraft fly". For trajectories, researchers could just use any of the other sites out there, but they wouldn't be able to, say, develop wind and temperature models [1] or analyze collision avoidance systems in depth [2]. It's the whole raison d'être of OpenSky. We are keenly aware it's not an option for everyone to feed OpenSky because of that, but it's not something we can compromise on. We rather have fewer feeders who can provide that sort of data instead of basically duplicating all the other pure tracking sites out there.
[1]
www.lenders.ch/publications/conferences/IPSN18_3.pdf
[2]
www.cs.ox.ac.uk/files/11077/DASC_2019.pdf