fix: shuffle to undo unique sort

[eddiebergman]

• Closes Neighbour Values of Integer Hyperparameter are not Symmetric #405

@LukasFehring Anyone going to take over on ConfigSpace from Hannover side? I'm not supported to really maintain ConfigSpace as I've left AutoML.org

[thijssnelleman]

@mfeurer I don't think I can contribute to this PR, it looks logical to me but I do not understand the full context.

[eddiebergman]

It's mergable, you can view the issue it's derived from and that it fixed the users issue.

Essentially, the sampling procedure would require doing a check "have we already sampled this value", which is much cheaper to check when sorted. However this sorting ended up biasing the sample, as we returned the first n samples, i.e. the first n samples according to order, instead of a random n samples around the distribution

[LukasFehring]

Sorry, for some reason i got no notification on this branch :)
The short answer is that I do not have the capacity to work on ConfigSpace

[thijssnelleman]

Alright, sounds good to me. @mfeurer I think with your approval this can be merged.