SparseArrays: specialize zero(...) so result has nnz(...) = 0

[tkluck]

This pull request fixes #31835.

Before this commit, the result of

zero(x::AbstractSparseArray)

is filled with stored zeros exactly where x has a stored value. That is a
wasteful artifact of the default fallback implementation for AbstractArray.
After this commit, we return a sparse array of the same dimensions as x
without any stored values.

This change is backwards incompatible where it relates to object identity.
Before this commit, for mutable objects, every stored zero has the same
identity. Example:

julia> using SparseArrays

julia> y = zero(BigInt[1,2,3]); y[1] === y[2]
true

julia> y = zero(sparse(BigInt[1,2,3])); y[1] === y[2]
true

julia> Base.zero(a::AbstractSparseArray) = spzeros(eltype(a), size(a)...)

julia> y = zero(sparse(BigInt[1,2,3])); y[1] === y[2]
false

But this behaviour should be classified as a performance bug and can therefore
be fixed in a minor (but not a patch) release.

[tkluck]

Thanks for approving @ViralBShah !

Just had a look at the CI failure. The only failure is buildbot/tester_win32 and nothing in the logs seems related to this change. Should be good to merge.