Properly implement floating point summation in FindShapes filter

The FindShapes filter calculates the second order moments of objects by summing over all the voxels belonging to a Feature. These values are on the order of 10^9, and since many summations occur, the floating point errors accumulate. This can be an issue for special cases where an eigenvalue of the moment matrix should be 0 but ends up being slightly larger or greater than 0. A proper floating point summation approach, such as Kahan summation, should be implemented.