Heavy hitter estimation is broken when there are multiple flows of the same size

[cppascalinux]

When getting the ground truth for heavy hitter, getHeavyHitter in data.h simply takes the largest $K$ flows, and use the size of the $K$'th flow as the threshold $thr$. However there might be multiple flows of the same size as the $K$'th one, and when testing, we ask the sketch to retrieve all flows of size $\ge thr$, which will definitely result in a drop in the recall rate. This effect is significant when the flow size are distributed evenly.

A possible fix could be including all flows of size $\ge thr$ when calculating the ground truth.