Fix fake GPS velocity frame and COG units

[Daniiiil1]

Summary

Fixes the fake GPS velocity and course fields reported in #2091.

Root cause

send_fake_gps() differentiated ECEF positions, reversed the displacement sign, and then wrote those ECEF components directly into MAVLink fields documented as North/East/Down. The HIL_GPS course calculation also produced radians but stored the result as centidegrees. The first sample additionally used uninitialized previous-position state.

Changes

• derive the ECEF velocity with the correct displacement sign
• rotate ECEF velocity into the local ENU frame and then into NED
• emit course over ground clockwise from north in centidegrees
• treat the first sample and non-increasing timestamps as zero velocity
• use the corrected NED velocity for both HIL_GPS and GPS_INPUT
• add focused tests for ECEF-to-NED axes, timestamp guards, and cardinal/intercardinal courses

Validation

• git diff --check
• added deterministic gtests covering north/east/up transforms and 0/45/90/180/270 degree courses

The ROS 2 toolchain is not available on the local macOS host, so the package test target is left for the repository CI matrix.

Closes #2091

[vooon]

@Daniiiil1 please fix cmake includes for this test.

[vooon]

Maybe worth to move to generic frame_tf.hpp?

[shupx]

@Daniiiil1 Great! Thank you for resolving this issue that has been troubling me for a long time. Prior to this, I could only work around it by writing a separate module to bypass MAVROS's fake GPS.