I accidentally "fixed" the reference solution for
test_2p2cni_waterair_buoyancy_box in !1932 (merged).
On BuildBot it fails however.
I know from many people that they have problems with this test. It doesn't seem to be very stable. We should have a better look at the test, identify what can be done to make it both a good test for the 2p2cni model and make it tractable for reliable testing.