AP_GPS: Fix undulation incorrect convention

[Ryanf55]

Purpose

Fixes #29199

Tests performed

Ground and flight tested on UBX, I would like to request help from the community on checking other GPS's.

TODO

• We should fix/clarify the comments in get_undulation and spread across the code. The language is either incorrect or imprecise.
• Get other users of GPS's to test this patch - you only need to power on your vehicle, get GPS lock, and send me a BIN and TLOG.
• Consider an autotest and starting to simulate MSL/ellipsoid differences in SITL
• Add documation to the dev wiki on conventions for writing a GPS driver?

[tpwrules]

Not tested yet. Needs some more sign flips:

ardupilot/Tools/AP_Periph/gps.cpp

Lines 101 to 103 in 6bf29ec

	if (gps.get_undulation(undulation)) {
	pkt.height_ellipsoid_mm -= undulation*1000;
	}

ardupilot/libraries/AP_OpenDroneID/AP_OpenDroneID.cpp

Lines 405 to 411 in 6bf29ec

	float undulation;
	if (current_location.get_alt_cm(Location::AltFrame::ABSOLUTE, alt_amsl_cm)) {
	altitude_geodetic = alt_amsl_cm * 0.01;
	}
	if (gps.get_undulation(undulation)) {
	altitude_geodetic -= undulation;
	}

ardupilot/libraries/AP_Mount/AP_Mount_Siyi.cpp

Line 1301 in 6bf29ec

position.alt_ellipsoid = position.alt_msl - undulation*100;

I wonder if we should rename dataflash GPA.Und to make it clear that the sign has changed for any tools which depend on it. Perhaps GPA.UndC for conventional, and possibly keep the old one? I would also support removing it and/or adding an ellipsoid height. Not sure dataflash compatibility is super important.

[tpwrules]

Tridge says:

• need to change dataflash name
• could log both, would prefer not to since the existing one is unexpected

NOVA driver might be wrong too, need to test

[tpwrules]

From page 245 of (an archived copy of) the NOVA protocol document om-20000129.pdf mentioned in the header:

The illustration matches Wikipedia and NMEA and is opposite the convention currently in ArduPilot, so the undulation received from the NOVA is almost certainly wrong. If so, this would make a pretty clear case that ArduPilot's convention is backward because it would make all drivers negate the quantity.