dief@t adds additional data point with wrong timestamp

[prohde]

The implementation of dief@t adds an additional data point for each approach with the maximum time t, where t is either the parameter given or the maximum t of all approaches. This leads to wrong values of dief@t if an approach finishes earlier than the others.

Assuming the following data

approach	answer	time
a1	1	379956750.0
a1	...	...
a1	2746	379966378.0
a2	1	61117396.0
a2	...	...
a2	2746	61117396.0

Approach a1 is less efficient but is generates answers continuously while a2 is a blocking approach. The current implementation adds the data points (a1, 2746, 379966378.0) and (a2, 2746, 379966378.0). Hence, the area-under-the-curve and the value for dief@t for a2 is higher than the one for a1 which is clearly wrong. The value of dief@t for a2 should be 0.0. This PR provides the necessary changes to fix this issue.

[prohde]

I discovered that the additional data point is necessary in most cases. However, when there is a true blocking approach as in the example above, it breaks the intuition. So I added a switch to turn off the behavior. By default, the data point will be added.