EKF2 global position validity

[bresch]

Solved Problem

1. Once in inertial dead-reckoning, the position estimate drifts exponentially and even if a velocity-aiding sensor starts after some time, the global position shouldn't be valid anymore.
2. FW manual position control mode requires a valid global position estimate but internally works using local coordinates. The mode thus cannot run without a GNSS fix or a manual reset despite having a "valid local position" (due to airspeed and sideslip fusion).
3. Flying in FW using wind dead-reckoning requires to set the COM_FS_EPH to a really large value as commander would otherwise invalidate the local position validity
4. FW auto takeoff and auto land modes don't require a valid global position to start but the backend can't execute correctly because it relies on a valid global position estimate.

Solution

1. Invalidate global position when exceeding maximum the inertial dead-reckoning time (5s). The global position can then be valid again if the local position is valid and that a global position measurement is available (e.g.: GNSS, AGP or manual pos aiding).
2. Make FW manual position work with local coordinates only
3. Generalize the local position "relaxed" logic (currently used for optical flow) to work when the local position is obtained from velocity-aided dead reckoning (e.g.: optical flow, ev_vel, airspeed+sideslip)
4. Deny switching to auto takeoff and auto land if global position is invalid. In the future we could also fix the logic to enable it correctly.

Note that we could publish NANs in the global position fields that are invalid but I wanted to follow the same scheme than in vehicle_local_position.

Test coverage

SITL tests:
In FW, start and stop several times GNSS and sideslip fusion. Modes switch correctly.

[Jaeyoung-Lim]

Yay! This is exactly why I removed global position from the guidance law 😁 Happy to see this finally getting used!

[mrpollo]

Really cool @bresch 👏

[dagar]

Does this even matter? Couldn't "relaxed local position validity" simply be vehicle_local_position.xy_valid?

[bresch]

Good point, I'll check if there is any reason why it should be different

[bresch]

I think it makes sense. Now "relaxed" is just without eph check.

[haumarco]

Right now, _NED_origin_initialised represents if the global_ref was set and the local position is valid. If the NED origin was initialized should be checked with _pos_ref.isInitialized() right? Maybe renaming it to _NED_origin_valid would be better, or would it be possible to remove it?