[fmrico]

Closes #166.

• projection parameter: utm (default, as before) or enu, east-north-up at the origin (GeographicLib LocalCartesian), for maps defined that way.
• The first fix places the robot and not the antenna: the antenna's mounting (TF from the filter's base_link_frame to the fix's frame) is turned with the IMU's attitude and removed. Later fixes carry the IMU's attitude, so the filter removes the mounting in the right direction. Without IMU or without that TF, as before.
• README: the GNSS parameters and the antenna.
• Tests for both projections, an unknown projection, the antenna at the first and later fixes, without its TF or without IMU, and the fix of the estimate published back.

[midemig]

Two things I’d change before merging:

1. The IMU’s mounting (TF) is not applied.
   update_rt uses IMUPerception.data.orientation as is, and every fix carries it. The IMU perception handler stores the raw message, so that orientation is imu_link’s, not base_link’s. With a mounted-turned IMU, the antenna offset is rotated with the wrong heading, on the first fix and on all later ones.

2. A null IMU quaternion gives NaN.
   An IMU with no orientation (orientation_covariance[0] = -1, quaternion 0,0,0,0) makes q.normalized() in antenna_to_robot return NaN. The first fix is NaN and the filter stays poisoned afterwards.

[fmrico]

Thanks, both were right.

1. The IMU's orientation is its frame's: the robot's attitude now removes the IMU's mounting (TF from base_link_frame to the IMU's header.frame_id). Later fixes also carry the antenna frame's orientation, which is what robot_localization uses to remove the antenna's mounting.
2. An IMU without orientation (orientation_covariance[0] == -1, a null or non-finite quaternion) or whose mounting is not in TF gives no attitude: the first fix places the robot at the antenna, with a warning, and later fixes use the filter's estimated heading. No more NaN.

d457a29, with 7 new tests (IMU and antenna mounted turned, no orientation, null/NaN quaternion, unknown mounting, no IMU, IMU appearing later). In the Prius playground with the IMU turned 90° and the antenna 1.5 m ahead, the error against ground truth was 4.4 m with a 90° heading error before, and is 0.16 m mean while driving now.