EKF: Improve robustness of yaw reset to bad inertial data

[priseborough]

Preliminary response to the corner case highlighted by https://discuss.px4.io/t/estimator-failure/18790

Testing to date limited to a basic test in SITL with the mag data rotated 180 deg to induce a flyaway, yaw reset and recovery sequence - https://review.px4.io/plot_app?log=b9381141-21c1-4401-af2f-4639690dea49

EKF-GSF convergence

Further testing is required including a way to simulate the clipping behaviour case encountered.

Another PR forthe PX4/Firmware ekf2 module to change setting of _ekf.set_in_air_status(!_vehicle_land_detected.landed) is recommended to ensure the EKF-GSF activates earlier.

[priseborough]

TODO - update comparison file for reversion diff test.

[bresch]

Ok, looks good now. However, why is this changing the output of the standard change indicator? If there is no emergency yaw reset, nothing should change.

[priseborough]

I can't explain that. Maybe it's the compiler version. I was building from a Ubuntu 20.04 machine when I updated the change indicator files.

Edit: Checking back to the CI test result before I pushed the update change indicator files https://github.com/PX4/ecl/runs/1208622495 shows that only the ekf_gsf_reset.csv file was generating a diff.

[priseborough]

There is a time delay between the landed flag going false and the EKF-GSF logging starting, see below. Thinking about this some more, some of the delay can be explained by the fact that the EKF-GSF yaw estimator is running on the same delayed fusion time horizon as the main navigation EKF loop which can introduce a delay of up to 200msec. A further 200msec can be explained by the 200Hz logging rate for EKF-GSF data and a further 200msec by the GPS update rate as the EKF-GSF won't initialise until the next fresh GPS sample is available, then has to wait a further 200msec for the next GPS measurement which explains the majority of the delay.

No further investigation of options for setting theEKF's in-air status earlier is warranted at this time.