Alexey Stukalov

@dmbates Unfortunately, I have discovered your [dmb/matrix_sparse](https://github.com/JuliaSparse/MKLSparse.jl/tree/dmb/matrix_sparse) branch, which overlaps with this PR, too late. If you have any chance to review, please let me know your suggestions.

@KristofferC Oh no, unfortunately, it's still there for the new `dense_matrix * sparse_matrix` on macOS/Julia-nightly as before (nightly on the other OS is fine); updating to MKL 2023.1 doesn't fix...

Actually, it would be nicer to use JULIA_MKL_LOAD_PATH="@stdlib/../../site" -- no need to fix *sysimg.jl* afterwards (though I don't know whether that is relevant at all). But at the moment `Base.load_path_expand()`...

> I still don't really understand why this is needed. Could you elaborate on what is special with ArchLinux that requires this? Until recently, the official ArchLinux Julia package was...

@mhauru I have incorporated your suggestions, thank you!

I have tried to tweak the layout of the plots and switched to SVG, let's see if that helps.

The updates did fix the dev docs.

Thanks for the PR! Supporting large datasets is definitely very interesting. We just have to be conservative with introducing additional dependencies to *Clustering.jl*. From the first glance, it doesn't look...

> All the changes are because the views function is not implemented for Arrow.Tables Still, if the only difference is the row access, you can define e.g.`_sample(D::AbstractMatrix, ix::Integer) = view(D,...

One problem that I see with `Arrow.Table` support after thinking a bit more is that it breaks the convention that samples are stored column-wise, because for the `Arrow.Table` they are...