Move negative metric values into lowest bucket

[JONBRWN]

What is the current behavior?

Currently, if there is a negative time difference between the receiving worker and the sending worker, latency metrics for the network time computation end up in the largest bin possible due to underflow.

What is the expected behavior?

In the ideal world, machines would be perfectly synced. Since this tends to not be the case from time to time, we should insert underflow metrics into the lowest bin as to not negatively skew the upper percentiles with a disproportionately large value that does not represent a real latency value.

This is currently only visible in the network time computation and should be addressed there. Updating other metric areas should happen only if this issue becomes visible elsewhere.