Definitely didn't crash. I note that it seemed to happen when flipping 180° and coasting backwards into the wind. The last time we had someone suffer weird issues like that was because idle throttle was too low, the ESCs started to desync and fuck up, because they didn't bring up enough power to counter the air blowing into the props trying to rotate them the other way. That might be worthwhile to check.
As far as the LQG propwash goes, for once, it might be that autotune has guessed slightly wrong.
LQG has two main components it uses for control, that rotational rate, which is an actual measured value, and torque, which is mostly modelled based on the autotune data. I figure that torque isn't always on point on certain extremer maneuvers, altho I rarely had any during testing of it. That might cause the propwash. It was a bitch to tune that thing to begin with.
If you really want to mess around with things, the Q1 and Q2 parameters in the LQRegulator tree of LQGSettings how much effort it is supposed to give trying to drive the control errors to zero. Upping it increases effort, lowering decreases effort. You might probably want to mess with Q2 a bit (FYI, they're not linear but quadratic, and are inputs to the LQR solver). R acts as a scaler for both. Alternatively, in the RTKF, its Q2 parameter controls how much it should use the gyro data to correct the torque estimate. Might probably want to mess with that, too. Probably upping it a little, which might however introduce some noise on the motors, if too high.
BLHeli and Betaflight introduced realtime telemetry on ESCs with the newest firmware. That can be exploited eventually, to guess actual torque values to supply to the RTKF. It might improve things, or it might not. I am meaning to implement this eventually, but the bit of time I'm currently spending on dR again is going to teeth gnashing as to why shit keeps randomly corrupting important internal structs with the RTOS upgrade I'm trying to make work.