provide non-aliasing constructors for sparse arrays

[KlausC]

This fixes #34630, which does not touch makes copy sparse constructors.

[andreasnoack]

I'm not completely up-to-date on the conversations on the sparse matrix constructors but around 0.6/0.7, we discussed the memory behavior of constructors and convert methods when the input is of the same type and I believe the conclusion was the convert should avoid copying when possible whereas constructors should avoid aliasing.

[KlausC]

convert should avoid copying when possible whereas constructors should avoid aliasing

this implementation of convert avoids copying, if possible (that means the argument has already the target type). If that is not the case, a type conversion is necessary, which implies at least one of the 3 vector fields of SparseMatrixCSC or the 2 vector fields of SparseVector has to be copied, too.
The demand, that the usage of published methods does not surprisingly destroy the integrity of some of the read-only arguments of those methods leads to this solution.

I agree with you, that it would be better to make copies in the constructor SparseMatrixCSC{Tv,Ti}(::AbstractSparseMatrx), but got backfire on this suggestion (see #34630).

[KristofferC]

I agree with you, that it would be better to make copies in the constructor SparseMatrixCSC{Tv,Ti}(::AbstractSparseMatrx), but got backfire on this suggestion (see #34630).

But the "constructor" for sparse matrices is sparse?

[KristofferC]

I do agree that it is pretty weird that the SparseMatrixCSC constructor is undocumented. Documenting it and making it the "first class" constructor would make sense to me.

Right now we have SparseMatrixCSC(M::Matrix) = sparse(M) but it should really be the opposite imo.

[KristofferC]

Deciding between doing this for convert and SparseMatrixCSC, I would prefer to copy for SparseMatrixCSC for the reasons outlined in #34630 (comment) (consistency with Array).

[KlausC]

Deciding between doing this for convert and SparseMatrixCSC, I would prefer to copy for SparseMatrixCSC

Ok, that is fine. I will change this PR accordingly.