spurious underflow in norm

[stevengj]

From discourse:

julia> using Colors, ColorVectorSpace, LinearAlgebra, FixedPointNumbers

julia> norm(RGB(0.008,0.008,0.008))
0.008

julia> norm(RGB{N0f8}(0.008,0.008,0.008))  # spurious underflow
0.0

We typically expect norm to avoid spurious underflow. It looks like you already do so for RGB{Float64}, since norm([RGB(1e-300, 1e-300, 1e-300)]) correctly returns ≈ 1e-300 rather than underflowing to zero.

I would suggest doing norm computation in the precision of the output (Float64) rather than in the precision of the input, as well as being careful about scaling.