Hello Romaric,
I tried your change. Attached you find a CSV file containing the output.
quat.csv (41.0 KB)
It looks like it switches the sign of the quaternion after the first 20 iterations and then again after 30 further transitions.
Hello Romaric,
I just took a look at my visualisation in Celestia. It actually looks quite good there. I use a step size of 0.1s between propagations and transition time 60s. AttitudeStateProvider was not changed. So it is now doing a transition that looks kinda realistic. If you found that there is a problem anyways please let me know. I will be happy to help you hunt it down on my side by offering my code as a testing ground 
Cheers,
Yannick