Can't differentiatate through StepRangeLen due to TwicePrecision

There is one very common data type that we can't differentiate through - StepRangeLen, which is what we usually get back from range.

julia> using ForwardDiff

julia> f1v(x::Vector) = sum(range(0; step=0.1, stop=2)*x[1]);  # edited not to clash with definitions below

julia> g1v = x -> ForwardDiff.gradient(f1v, x);

julia> f1v([1.0])
31.5

julia>  g1v([1.0])
MethodError: no method matching twiceprecision(::Base.TwicePrecision{ForwardDiff.Dual{ForwardDiff.Tag{typeof(f1),Float64},Float64,1}}, ::Int64)
Closest candidates are:
  twiceprecision(!Matched::T<:Union{Float16, Float32, Float64}, ::Integer) where T<:Union{Float16, Float32, Float64} at twiceprecision.jl:220
  twiceprecision(!Matched::Base.TwicePrecision{T<:Union{Float16, Float32, Float64}}, ::Integer) where T<:Union{Float16, Float32, Float64} at twiceprecision.jl:225

This problem does not arise with numerically equivalent data types which do not use TwicePrecision, such as LinRange:

julia> using ForwardDiff

julia> f2(x::Vector) = sum(LinRange(0, 2, 21)*x[1]);

julia> g2 = x -> ForwardDiff.gradient(f2, x);

julia> f2([1.0]), g2([1.0])
(21.0, [21.0])

Related: https://github.com/JuliaMath/Interpolations.jl/issues/293

Start by reproducing the ForwardDiff.gradient example using range and compare it with the working LinRange case. Then inspect how StepRangeLen and Base.TwicePrecision are handled during differentiation, using the reported twiceprecision method error as the guide; done means the range example differentiates successfully like the LinRange example.