Optimize get_dataset_response and optimal binning

get_dataset_response was pretty invasiv in the users
setup (most would not know, but as soon as you start
fiddeling with assigned models you would be hit).

Now we copy a dataset over where we assign a function
for the delta peak and evaluate here. After deleting
the user has its environment back.

Also streamlined how the response is calculated, allows
now to select a specific channel instead of getting the
full response, which might be huge.

Using this, the memory footprint of optimal binning
was reduced, resulting in significant runtime
improvement.