As can be confirmed by checking generic code for this operation, it is supposed to blend between the result of Breakdown based on actual frame range, and the current pose. However for some reason the quaternion specific code was blending between the current pose and the current keyframed pose. This means that the operation does nothing if invoked without modifying the pose first. This rewrites the code to match the non-quaternion behavior. Differential Revision: https://developer.blender.org/D13030