Scalar indexing on `+(::Symmetric, ::Symmetric)`

[jonas-schulze]

The Minimal Working Example (MWE) for this bug:

using CUDA, CUDA.CUSPARSE
using LinearAlgebra
using SparseArrays
using Test

n = 10
A = sprandn(n, n, 2/n)

@testset "$T" for T in (
    CuSparseMatrixCSC,
    CuSparseMatrixCSR,
    CuSparseMatrixCOO, # broken as well, but not important to me
)
    global Ad = T(A)
    @test_broken Ad' + Symmetric(Ad) isa T # scalar indexing
    @test_broken Ad + Symmetric(Ad) isa T # scalar indexing
    @test_broken Symmetric(Ad) + Symmetric(Ad) isa AbstractMatrix # scalar indexing
    # ...ideally, the last should be a Symmetric, though.
end

Project.zip (same as #3042)

Expected behavior

It should just work.

Version info

Same as #3042.

[kshyatt]

Just in general I think our sparse + any other array wrapper type stuff is quite janky, so thanks for looking into these annoying intersections