GPU implementation of hamming distance

[felixpetschko]

Hamming distance implementation with numba.cuda for GPU support.
This is built on top of the changes in Hamming distance implementation with Numba #512

@grst:

• add to API docs
• start small "large dataset tutorial", see also Large dataset tutorial #479
• update changelog

Close #479

[Intron7]

For a large Matrix in the Hamming-Kernel have you checked that int is enough to cover the indexing of data?

[felixpetschko]

For a large Matrix in the Hamming-Kernel have you checked that int is enough to cover the indexing of data?

That's a good point, but I think for our purposes int should be enough for the kernels. I think the most problematic part is the indptr array of the CSR matrices, therefore I added some int32 overflow checks. If such an overflow occures, the user can try it with a smaller cutoff to get less result values that need to be stored in the CSR matrix.