ADAlgorithm: New SI does not always start with best controls

[Ferlemann]

An example can be seen in the plot:

Origin of the issue: If during an SI no new best control parameters are found, the best control parameters of the previous SI are reused (in combination with the new basis), so that the base pulse is modified (arbitrarily) while it should remain untouched.

The same issue could probably be seen for the GRAPEAlgorithm.