Document that force-torque sensor should only be attached to joints, not links

It has always been my understanding based on the [usage in osrf/gazebo](https://github.com/osrf/gazebo/blob/6fd426b3949c4ca73fa126cde68f5cc4a59522eb/test/worlds/force_torque_frame_test.world#L88-L163) that a sensor of type `force_torque` may only be a `//joint/sensor` and not a `//link/sensor`. I believe it is not stated in the specification, and the [test/sdf/sensors.sdf example](https://github.com/osrf/sdformat/blob/sdformat10_10.0.0/test/sdf/sensors.sdf#L140-L142) even includes a `//link/sensor` of type `force_torque`. I noted this while reviewing the new `sdf::ForceTorque` DOM element, which does not enforce this condition, in keeping with the pattern currently set.

I would propose that we formalize this in SDFormat 1.8 and generate parsing errors if a `//link/sensor` has type `force_torque`.

Start with the referenced test/sdf/sensors.sdf example and the gazebo force-torque sensor test, then inspect the new sdf::ForceTorque DOM element and its parsing path. Done means the SDFormat 1.8 specification states that force_torque sensors belong only under joints, the conflicting link example is addressed, and link-attached sensors produce parsing errors.

Written by the indexing model from the issue text.