Distribute logic to multiple specialized ROS nodes to improve maintainability

[hmakelin]

Refactor from a single multiprocessing Python node to a more distributed system that should be easier to maintain:

• AutopilotNode abstract base class adapter for given autopilot (PX4Node/ArduPilotNode)
  • Provides a stable internal interface to autopilot data, removes old gisnav.autopilots subpackage
• BBoxNode publishes a BoundingBox for approximate location of camera field of view (FOV) projected to ground
  • Old FOV guesstimate logic moved to dedicated node
• MapNode retrieves orthoimagery/DEM for the BoundingBox and publishes a custom OrthoImage3D message (http://github.com/hmakelin/gisnav_msgs)
  • Holds WMS client inside
  • The OrthoImage3D message is needed to enforce atomicity of the orthoimage and DEM pair, they need to align as well as possible
• PoseEstimationNode estimates GeoPose between camera image and orthoimage/DEM
  • Holds the neural network, also does some OpenCV
  • Gets the geographical bounding box data inside the OrthoImage3D message so can estimate GeoPose and not just Pose
• MockGPSNode generates and publishes a mock GPS message based on the estimated GeoPose
  • Transforms GeoPose (and complementary Altitude) message to a mock GPS message such as PX4's SensorGps
  • No separate nodes for PX4/ArduPilot unlike with AutopilotNode (see px4_micrortps ROS parameter)

[hmakelin]

Initial refactoring into multiple ROS nodes is now done. Will next focus on cleaning everything up. This is a major PR so will also update the documentation before merging.

[hmakelin]

Update: Refactoring should be done, now focusing on testing and updating documentation.