Prevent shutdown ordering crashes

Too buggy to leave it in for 3.0 release

- Also fix vehicle selector dropdown in toolbar

- Pervasive change to automatically connect to known board types
- Lots of additional changes along with this:
- More exposure of objects/properties to Qml
- Much restructuring in LinkManager and LinkManager clients

- Pervasive change to automatically connect to known board types
- Lots of additional changes along with this:
- More exposure of objects/properties to Qml
- Much restructuring in LinkManager and LinkManager clients

- Pervasive change to automatically connect to known board types
- Lots of additional changes along with this:
- More exposure of objects/properties to Qml
- Much restructuring in LinkManager and LinkManager clients

Instead us a Toolbox concept which hangs off of QGCApplication

- All resources are either under src or under resources dirs
- Removed files top level directory
- Move flightgear files to top level dir

Link ids were being used interchangeably with the mavlink channel for
the link. Link ids handed out as a numerically increasing integer. Once
the counter went past MAVLINK_COMM_NUM_BUFFERS all hell would break
loose and cause memory corruption.

Also many LinkManager API changes to further isolate all link
management inside LinkManager.

Moving methods from MainWindow to QGCApplication so that more things
can run without a window. Helpful for testing.

Plus a large number of other changes to allow for orderly shutdown of
objects without crashes or asserts. This is need for unit tests to
create/delete global state around every test.

Using QFileDialog directly on OS X can hang intermittently due to Qt
bug.

- Logging now always on to temp file
- User prompted to save temp file after disconnect
- Orphaned temp logs checked for on boot
- Logs saved to new save file location
- Log file starts/stops with first and last connnection

In the logplayer, the data stream is read ahead for messages, which is used to determine the size of the output bytearrays. Previously, this wasn't done correctly resulting in non-ideal behavior. Now, the next message is scanned properly, so the only bytearrays that are emitted from QGCMAVLinkLogPlayer are MAVLink messages in their entirety. This also has the benefit of skipping invalid/corrupted messages in the log that's being played back.

Previously MAVLink data streams recorded by QGC would use a fixed block size of the maximum MAVLink message length and fill in only the bytes written by the message. This wasted space, make manual parsing difficult, and broke compatibility with scripts provided in the MAVLink project (issue #174).

This patch alters logging to output only a packed data stream (64-bit big endian unix timestamp in microseconds since epoch + MAVLink message) instead of the unpacked data stream previously output. Additionally the previous logging code used the system endianness for packing in the timestamp, this has now been switched to always be big endian regardless of platform. All the documentation specifies big endian, so the code now follows the docs here.

Additionally data playback has been modified to playback both the new packed data streams as well as the old data streams, even those with improper endianness for their timestamps.

Finally, a variety of bugs have been fixed, along with some additional features and user experience changes, hopefully for the better. All existing functionality has been preserved as well.