for passive RTT measurements with QUIC, I would usually recommend using the latency spin bit, which was introduced for exactly this purpose. It has various advantages compared to parsing acknowledgments. Unfortunately, quicly (the QUIC library behind qperf) does not support the spin bit, so if you don't want to implement it on your own, you have to decrypt the pcap and parse ACKs.

I don't fully understand your argument on why you can't use the "Largest Acknowledged" value. In QUIC, packet numbers are acknowledged. The packet number space differs for each direction, but this does not affect RTT calculation. In theory, it should give you some samples if you look at the timestamps of packet numbers and the timestamps of corresponding ACKs.

You should also keep in mind that in QUIC, acknowledgments are delayed, and the values you retrieve are, therefore, very likely higher than the actual RTT.