When publishing the point clouds for visualization, for each LiDAR frame, we unnecessarily transform the local map points to the sensor frame to publish it with the same frame as the LiDAR frame. This PR avoids the transformation by directly publishing the local map points in their frame_id.

We already discussed this in #285. After #371, in which we added the odom_lidar to the tf tree, we can now easily publish the local map at this frame_id (assuming that publish_odom_tf is set to true). If publish_odom_tf is false, we will now see both the estimated trajectory and the local map because both are published in odom_lidar, which is the default fixed frame in rviz.

This is also how we currently do it in Kinematic-ICP.