derivative(f!, y, x) doesn't work if y is complex, even if x is real

Here is an MWE. For f(x) generating a complex array for x::Real, derivative(f, x) works:

julia> VERSION
v"1.12.6"

julia> using ForwardDiff: derivative

julia> f(x::Real) = [cis(x), cis(x)];

julia> derivative(f, 0.0)
2-element Vector{ComplexF64}:
 -0.0 + 1.0im
 -0.0 + 1.0im

However, derivative for an overwriting f! generates an error, if f! overwrites a complex array:

julia> f!(y::AbstractVector{<:Complex}, x::Real) = (y .= cis(x));

julia> y = Vector{ComplexF64}(undef, 2);  # f!(y, x) produces same result as y .= f(x)

julia> derivative(f!, y, 0.0)
ERROR: MethodError: no method matching ForwardDiff.DerivativeConfig(::typeof(f!), ::Vector{ComplexF64}, ::Float64)
The type `ForwardDiff.DerivativeConfig` exists, but no method is defined for this combination of argument types when trying to construct it.

Closest candidates are:
  ForwardDiff.DerivativeConfig(::F, ::AbstractArray{Y}, ::X) where {F, X<:Real, Y<:Real}
   @ ForwardDiff ~/code/ForwardDiff/src/config.jl:82
  ForwardDiff.DerivativeConfig(::F, ::AbstractArray{Y}, ::X, ::T) where {F, X<:Real, Y<:Real, T}
   @ ForwardDiff ~/code/ForwardDiff/src/config.jl:82

Stacktrace:
 [1] derivative(f!::typeof(f!), y::Vector{ComplexF64}, x::Float64)
   @ ForwardDiff ~/code/ForwardDiff/src/derivative.jl:27
 [2] top-level scope
   @ REPL[6]:1

Is this an intentional limitation of derivative(f!, y, x)? Given that f and f! in the above example perform the same calculation except that the latter is overwriting, it looks strange that derivative(f!, y, x) doesn't work while derivative(f, x) works.

Start with derivative.jl and config.jl, especially the derivative(f!, y, x) entry point and the DerivativeConfig methods shown in the error. Compare the real-array constraints with the complex output example, then determine whether the limitation should be fixed or documented and validate the result with relevant coverage.