- pusher code now works with the new manual setpoint generation

Instead of publishing before even having done update(), let's wait for a
result and then advertise.

This contains the latest fixes for the SDK 3.0.

Note: {-1} initializes only the first element, and sets the others to 0.

This had multiple issues:
- linked list was modified while other instances were still running and
  accessing it (the used linked list is NOT thread-safe).
- Mavlink instance was deleted, but it was still in the linked list, and
  thus could still be dereferenced by other threads
- the instance was deleted, but it was still accessed by the 'stop-all'
  calling thread

What we do now is:
- wait for all threads to exit
- then remove the instances from the linked list and delete them

This reverts commit cefc00af.

* Added more GPS checks for LPE.

* Add logic to force GPS alt init.

we now use the already existing buffer for logging messages, which is
allocated on the heap.
In fact, stack usage was too high before this, now it's ok again.

Also, it's not clear where the allocation was. It looks like it was on
the heap, but the compiler could decide to put it on the stack. This is
very bad for us because we use fixed size stacks with tights bounds. So if
a user specifies a large topic to log, it could have crashed.

Now the allocation is on the heap and the user can specify any size of
topic to log (as long as there is enough memory).