sampling: implement continuous sampling of time and temperature values

[jonaspleyer]

Currently a grid search is performed with a random sampling method and multi-start initial values.
Add additional sampling method that uses continuous times but has a penalty whenever times are not matching defined time steps.
Implement it in such a way that in general only multiple of a smallest time step dt are not penalized. This way, we retain generality.

Use a multi-start with possibly random sampling or something else.
After that, iteratively do some form of gradient optimization.