Formulation of "gradient limit" needs rework

[p-snft]

The formulation of the gradient limit is at least hard to follow. I think I did not alter it in #1077 but @jokochems, ,who changed it before, now means it smells (see #1077 (comment)). If there is an issue, it will become obvious at latest with #1080.

At the same time, it also needs rework as it does not consider the lenght of the time steps but limits the delta between two time steps. This means, it is not possible to have a constant ramp in non-equidistant time grids.