Maintain a list of address/port pairs instead of a address/port map.

* Flight modes still not implemented. Waiting on message change.
* Also includes various cauterization of UI things which should not be
available on high latency link.

This ensures no ports are being displayed to the user that are definitely unusable.

The earlier sleep calls disabled the app event queue, leading to "beachballing" in Mac OS and a visible boot delay / jitter. Running the event loop actively while waiting prevents the UI from freezing.

We were connecting to system peripherals before, which adds boot delay as QGC re-tries opening the port. We were also opening serial ports two times on Mac OS: Once as /dev/cu.usbmodem1 and once as /dev/tty.usbmodem1 - since the cu.usbmodem alias was already open, the 2nd opening always failed, leading to further boot delay.

Fix TCP disconnection problem.

Still does not support multi-vehicle

Fix merge issues

This reverts commit 52160d80.

This is important to avoid having QGC hijack links that are active (we read 2000 bytes!) but are definitely not MAVLink. This can have nasty side-effects, e.g. if we would talk to an LTE modem in a laptop or alike.

This extension introduces a complete version handshake mechanism which will detect wether the vehicle AND the complete link / routing infrastructure support MAVLink 2. If they do, QGC will switch to the highest protocol version supported by all connected vehicles. If a new vehicle connects, it will re-negotiate the highest possible version. This means that we error on the side of compatibility, which later can be easily changed.

Remove/comment off debug output pollution

Pulling mavlink with a reversal for the CAMERA_IMAGE_CAPTURED message. It's now back as it was so it won't break existing code.
The pull also brought changes to the GPS_RAW_INT message. Only MockLink uses. I entered zeros for all new fields.