Bootstrapping uncertain datasets

[kahaaga]

As suggested in #24, it would be nice for the user to plug in various bootstrapping resampling methods. This is not very interesting for resampling single uncertain values (except perhaps in the case of equiprobable populations; see #41), but may be very relevant when sampling uncertain datasets.

I'm not sure how this should be done yet, but we need a separate interface for
subtypes of AbstractUncertainValueDataset and for UncertainIndexValueDataset.