mc_att_control: adjust attitude setpoint upon estimate reset

[MaEtUgR]

Describe problem solved by this pull request
When the attitude estimate quaternion is reset the attitude controller uses the new reference in the next loop iteration. But the attitude setpoint is not updates to the new reference until the position controller generates a new attitude setpoint but it runs in a much slower rate. As a result there are glitches in the output rate setpoint. See #14676 (comment)

Describe your solution
The internal setpoint copy is updated with the delta quaaternion upon attitude estimate reset. It will get overwritten with the next attitude setpoint change.

Test data / coverage
Completely untested, tests still REQUIRED.

[MaEtUgR]

http://ci.px4.io:8080/blue/organizations/jenkins/PX4_misc%2FFirmware-SITL_tests/detail/PR-14738/3/pipeline
If I test VTOL manually on my local machine I see no issue. Let's try again.

[MaEtUgR]

@dagar During SITL testing I found the following case is quite common: A new attitude setpoint that was generated with the old reference overwrites the old adapted one because they get processed in the same loop iteration. Easy fix, first check for setpoints then for resets -> 3rd commit.

Doing above I found the far less common but still occuring case where there's a new attitude setpoint that was already generated with the new reference and it gets corrected again assuming it's an old one (see above). Since this is less likely I prefer the solution with the fix but it's annoying to not have all corner cases covered.