feat: Resampling saves param vals during instantiation

[sebffischer]

Some inference method that e.g. provide confidence intervals for the generalization error need access to these parameter values. Currently we had to trust the parameter values are not changed after instantiation, which is suboptimal.

[mllg]

If the only purpose is to protect paramsets from modification after initialization, there are better ways (make param_set an active binding).

[sebffischer]

It is about instantiation of the resampling, and not initialization. E.g. when analying a benchmark result, I want to know which resampling parameters were used during the benchmak() call. But in principle it is possible to modify the $values slot of the parameter set after the benchmark call.