Improve duration comparision for "fast" workflows.

[mile95]

Currently, there is something off with the comparison.

This is mainly visible for very fast workflows.

It seems that the historical durations which we fetch using the GitHub API are much longer compared to the current run (which we measure ourselves).