Stop mag heading fusion if yaw stays at +/-90 deg

[kamilritz]

While I wrote a unit test, I noticed that when we set the orientation exactly to +/- 90 degrees that the heading uncertainty is increasing until it hits the limits. I followed it down to these lines of code. Since in practice we will never stay exactly so close to +/- 90 degrees we will not always return before triggering the fusion. But if we do, we stop performing heading fusion.

@RomanBapst @priseborough FYI

[mhkabir]

I wonder if this can also happen if you're using external vision which initializes to such an heading angle.

I still have't conclusively tracked down #698, and then there also was someone else reporting this: PX4/PX4-Autopilot#13116.

I am increasingly suspicious. Let me post a log.