BallJoint torque control applies AMotor torques to the ball joint instead of the AMotor

[omichel]

Spotted while reviewing #7028.

In WbBallJoint::prePhysicsStep(), the torque-control branch (userControl()) calls:

dJointAddAMotorTorques(mJoint, -rm->rawInput(), 0.0, 0.0);

It does the same for rm2 and rm3. But for a BallJoint, mJoint is the dBallJoint created in setJoint(), not the AMotor (mControlMotor).

• In debug builds, this trips checktype() in dJointAddAMotorTorques().
• In release builds, the ball joint's memory is read as a dxJointAMotor, so the applied torques are undefined.

Expected: wb_motor_set_torque() on a BallJoint motor applies the torque about the corresponding AMotor axis.

Notes for the fix:

• Pass mControlMotor instead of mJoint.
• dJointAddAMotorTorques() asserts that the AMotor isn't reversed (dJOINT_REVERSE). That's exactly the case when the BallJoint's parent solid has no physics, so it needs separate handling.