Support for DOK slicing and advanced indexing

Since the DOK array is versatile applicable I use them frequently in my projects. I needed support for slicing and advanced indexing and implemented it myself, see gist.

The indexing supports slicing, advanced (fancy) indexing and as I need it most of the time "array indexing". This is a fast version for advanced indexing. Only vectors (1d sequences) are accepted for indexing. The getitem method always returns dense ndarrays in contrast to normal advanced indexing of sparse arrays which create new copies of sparse arrays. The setitem method allows only vectors and scalars as values. This allows fast calculations over ndarray methods without the expensive creation of temporary sparse arrays between operations. This is of course only useful for small vectors, which fit easily in to memory.

If there is interest in adding this to pydata.sparse, I could help. But as I have never contributed to any open project, I have no idea if I could handle the necessary amount of work.

Review the linked sparse_dok_indexer.py gist and the DOK getitem/setitem implementation. Compare its slicing and advanced-indexing behavior with the project's current indexing conventions, then define coverage for vector indices and scalar or vector assignment; done should match the stated dense getitem and supported setitem behavior.