Move findnz to SparseArrays module

[nalimilan]

Its definition (return stored entries) only makes sense in the context of sparse arrays. Closes #24910. Part of #10593.

The first commit should be kept separate, it fixes an inconsistency spotted at #24910 (comment).

[mbauman]

No need for summing count anymore, either.

[nalimilan]

Indeed. I copied the SparseMatrix method too litterally.

[Sacha0]

lgtm! Thanks @nalimilan! :)

[Sacha0]

Tangentially, the more I consider this function, the more convinced I become that Stefan's "kill it with fire" quip was spot on. What particularly bothers me is the difference in this function's behavior between dense and sparse arrays; in other words, I am not certain I could write a generally valid docstring for this function as it stands. Best! :)

[nalimilan]

Tangentially, the more I consider this function, the more convinced I become that Stefan's "kill it with fire" quip was spot on. What particularly bothers me is the difference in this function's behavior between dense and sparse arrays; in other words, I am not certain I could write a generally valid docstring for this function as it stands. Best! :)

Yes, I was actually going to say something similar. It would make sense to deprecate the non-sparse methods in favor of find!(iszero, x), since that's exactly what they do (though they return the indices in a slightly different form). That way the difference of behavior with the sparse methods would be clear. We don't support e.g. nnz for non-sparse arrays.

[JeffBezanson]

It looks like the doc string for findnz should be improved to clarify that it returns all stored entries.