Instead of always starting with instance 0 (potentially an internal
mag), first take the current sensor as reference to compare the other
ones against it.
The issue is that otherwise we can end up in a
situation where a switch occurs because the currently used sensor isn't
much better than a sensor with a lower index: because the selected one
isn't much better, we cannot "fail-over" to it and get stuck on another
sensor, triggering a fail-over.
bresch