In the case of the compass calibrator we do not want to use the GSF
result if any model is degenerate. We've had a compass calibrate in
flight 180-degrees out from what it should have.
This improves protection against a condition where if all yaw hypothesis filers have very large innovations, the weights of Gaussian densities can underflow leading to failure of the weights to converge.
The setting of EKF state variances is only required when commencing or recommencing velocity fusion.
The function that resets the EKF and GSF class variables has been renamed to be more consistent with its function.