Treat Fork Handle as a Free Joint

[amalnanavati]

Currently, we treat the fork handle as a rigid joint. However, that is not quite the case. Because the base of the arm (where it attaches) is rubber, and because the fingers have some small movement ability, the fork handle is able to rotate slightly in botht he pitch and yaw directions. As a result of that, the real forktip is sometimes not at the forkTip frame. This creates issues for: (a) skewering the center of a bite in acquisition; (b) accurate food-on-fork detection (#169 ).

This PR is two-fold:

1. Add an Apriltag at a known location on the fork handle. Update the URDF to treat it as a free joint. Then, publish values for that joint based on the detected Apriltag. See [ada_imu](https://github.com/personalrobotics/ada_ros2/tree/main/ada_imu) as an example.
2. Once that is done, recollect ROSbags and retrain the food on fork detector, so that we can be certain that it is detecting deviations with respect to the correct fork pointcloud.