Fix warnings in second order tests

[gdalle]

They are caused by EnzymeAD/Enzyme.jl#1354, because we define our array-to-scalar scenarios as:

f_vector_scalar(x::AbstractVector)::Number = sum(sin, x)
f_matrix_scalar(x::AbstractMatrix)::Number = sum(sin, x)

If instead we do the following, problem solved, but it allocates:

f_vector_scalar(x::AbstractVector)::Number = sum(sin.(x))
f_matrix_scalar(x::AbstractMatrix)::Number = sum(sin.(x))

We need scalar-valued functions which

• do not allocate
• do not use LinearAlgebra
• have different derivatives for every component

My suggestion would be

function f_vector_scalar(x::AbstractVector)::Number
    s = zero(eltype(x))
    for i in eachindex(x)
        s += sin(i * x[i])
    end
    return s
end

but it causes JuliaDiff/Diffractor.jl#280

[gdalle]

Disappeared for now because we're no longer testing that specific combination