This is what the `quantile_newton()`/`cquantile_newton()` does,
because otherwise they were able to end up in an endless loops,
when the initial point and the mode are the same, see JuliaStats#666.

I'm not sure this is needed here, but the next change
is going to refactor them to use general `newton()`,
which would make this change anyway,
so unless we need the current behaviour,
let's do this change explicitly.

…action,

…the `newton()`

… the `newton()

These have `df(x)=1`, at least that is what the `df(x)` is if we solve
`Δ(x)=f(x)/f'(x)` as an equation for `f'(x)`.

Note that we really do need `Roots.jl` 2.2.0-or-newer,
because of JuliaMath/Roots.jl#445

From the disscussion in the PR, it is not at all obvious to me
what the proper solution here is.