Update tolerances for starting and ending position

[ted-miller]

Originally from here: #233 (comment)

believe that Ros_Controller_IsInMotion() is not strict enough. Specifically START_MAX_PULSE_DEVIATION should be decreased. (Or make a new constant for "END deviation".) (Maybe make this configurable in yaml?)

A brief test indicated that the fb pos trails the cmd pos by about .0009 radians. If my calculations are right, for some axes (such as GP12 T axis) that's only 24 pulse counts. So, once the increment-queue is empty, Ros_Controller_IsInMotion would immediately return false, even though the robot is still moving. [This could cause the robot to think it's finished while outside of the final tolerance position.]

I'm also suspicious that the difference between fb and cmd could shrink as the robot is decelerating. This might make it easier for other axes to also exceed the threshold.