Change xindexview to take multiple xarrays instead of vector of xindex

[wolfv]

We could change xindexview to work with vectors of indices instead of a vector of xindex, as numpy does. I.e. instead of taking std::vector<xindex>, we could take a number of xt::xarray<size_t> vectors, one per dimension.

Then, when resolving to the index, we just have to multiply with the corresponding stride, which should be fast in e.g. the assignment loop.
We could also store the offsets instead of the vectors, and calculate a shape and reuse the xstridedview to also resolve cases where the number of supplied vectors is lower than the dimension.
However, it feels that this would come at a cost for the case where scalar elements are selected.