[feat] Integrate camera information to local SLAM

• A new package is used to read camera data : cv_bridge. It belongs to the full desktop ROS opensource modules.

• The synchronization of the Lidar and the camera is performed using the closest camera frame from Lidar timestamp. This supposes the Lidar and the camera are fully synchronized and the data are available at the same time (no process is applied to the raw data).

• Pixel matches are built using OpenCV sparse optical flow. (The dense mode is tested in one commit but it is not efficient as we use only some pixels)

• Some bad matches are rejected using specific heuristics

• A new Ceres constraint is created with a pixel match using reprojection of the 3D points. Residuals are added to the optimization from the remaining matches.