normalize, width = true about fhist.jl

[Moelf]

so we often have histograms with "overflow" bins; since you don't have bin width information there, you can't take integral with width, thus can't normalize with width=true.

I do agree in statistics width=true makes most sense outside of high-energy physics.

[alanedelman]

seems like a tail wagging the dog thing to me
perhaps have width = unknown or something for overflow

[Moelf]

I see two solutions:

1. when dealing with overflow histograms, force width=false, hope user reads the docs
2. use the width of the overflow bin "as is", hope the approximation is what user wants.

[Moelf]

as you can see width is an explicit option one has to add to in the C++ Framework used wildy in the field:

https://root.cern/manual/histograms/#normalizing-histograms

let me try a PR idea

[Moelf]

https://juliastats.org/StatsBase.jl/stable/empirical/#LinearAlgebra.normalize

maybe we should adopt this interface (more work)