Add an `AbstractShiftedProximableNLPModel`

[MaxenceGollier]

In RegularizedOptimization.jl, we often construct subproblems of the form

$$\min_s \varphi(s; x) + \psi(s; x) + \frac{\sigma}{2} \lVert s \rVert^2$$

where $$\varphi(s; x) = f(x) + \nabla f(x)^T s + \frac{1}{2} s^T B(x) s$$ and $$\psi$$ is a shifted model of a regularized $$h$$.

Just like RegularizedNLPModel wraps both an NLPModel and a regularizer h, I suggest we add an AbstractShiftedProximableNLPModel <: AbstractRegularizedNLPModel that wraps a shift of a RegularizedNLPModel.

Related: JuliaSmoothOptimizers/RegularizedOptimization.jl#316.