add correct hardware interface for ros control

[jacknlliu]

• move the subscriber about the non-realtime read state from KUKA real-time controller to a standalone thread (ros callback of the subscriber is realtime incompatible, it interrupts the ros control loop). (the current joint position topic is about 25ms delay in ros control node, but the rospy rostopic just 5ms, it's confusing...)
• also for the java side, not put the read state from RT controller(it's very slow) in the real-time loop thread.