separate level is incorrectly assumed to be atomic

[willow-ahrens]

@wraith1995

    let
        A = Tensor(Dense(SparseList(Element(0.0))), fsprand(UInt, 42, 42, 0.1))
        x = Tensor(Dense(Element(0.0)), rand(UInt, 42))
        y = Tensor(Dense(Separate(Element(0.0))))

        @finch begin
            y .= 0
            for i = parallel(_)
                for j = _
                    y[i] += A[i, j] * x[j]
                end
            end
        end

        @test norm(y - A * x) / norm(A* x) < 1e-10
    end
    ```
    
    This should raise a concurrency error, but it does not