This should fix the regression where the mag values were not properly initialized and caused EKF2 to stop in CI.