this is more consistant and should enable it on qurt/linux

This makes sure the driver fails if the device path is invalid (::open fails)

mainly the default device path doesn't work

superseded by src/templates/module

Just to make sure it's built with CI

Comment was a contradiction

Instead of using a timeout to resend a mission item, we should directly
send the mission request again. This is needed because we removed the
automatic resending which lead to troubles.

if anything doesn't go according to protocol we should not just throw
warnings but actually tell the ground station with a mission nack.

When we receive a MAV_MISSION_UNSUPPORTED, we should try to send a
mission request again with or without int mode.

This is an attempt to fix mission upload and download with QGC on a
lossy link. The way it should work according to the protocol on
https://mavlink.io/en/protocol/mission.html is that in general
the ground station should be concerned about timeouts and
retransmission. This means the autopilot does not need to retransmit by
itself but just answer requests.

This seems like it would make the transfer less robust, however, it
actually resolves an edge case where QGC fails to make requests because
it gets spammed by the autopilot and keeps resetting the timeout.

This is important for the compile scope.

heading towards ROI: do not require setting the MIS_YAWMODE param to enable multicopters yawing towards ROIs

ROI yaw control: configure yawing capability of mount in vmount parameters.
If configured that mount has no yaw control, vehicle will yaw towards ROI, irrespective of MIS_YAWMODE.
Vehicle will behave according to MIS_YAWMODE when there is no ROI.

or receiving a command that we need to handle. This makes sure that we
don't drop commands, when the output mode is set to mavlink (in that case
the input topic is rate-limited, and each input change results in a
vehicle_command publication, which results in queueing up orb items if we
get vehicle_commands from other sources too).

It causes missed input changes.
This reverts 73d23742.

This reverts commit b7404ad6.

The way the broadcast IP is currently fetched is by sending an ioctl()
request. The limitation of this is that the broadcast address may not
be set on the network interface (more specifically, it is usually not
set in docker containers). In those cases, it results in the broadcast
address becoming 0.0.0.0, which is not valid [1].

This fix uses the network IP and the netmask to compute the directed
broadcast address. This should be more reliable, as both of those are
always set on the network interface.

[1]: https://tools.ietf.org/html/rfc1122, section 3.2.1.3 (a)

* Cygwin: repaired NuttX build after 1f63d858
all the NuttX specific WINTOOL define handling was skiped in cmake Make.defs generation
The Nuttx build is makfile based and needs its cygwin treatment

Additionally the ${PX4_SOURCE_DIR}/src/include which was added through cmake needs path conversion

The two root causes for the special handling are:
- ARM GCC for Windows doesn't support cygwin paths passed as an argument
  so they need to be either relative or converted via cypath tool
- Symbolic links are totally different under Windows and because NuttX uses them extensively
  it has special handling scripts that need to be fed with the correct defines

* Cygwin: NuttX include paths all converted in Make.defs.in

differentiate between WINTOOL define (MSYS & cygwin) and cygwin using uname command:
- MSYS needs symbolic link handling for ARM GCC but no path conversion
- Cygwin needs both