Allow for sparse Hessians in solvers

## What do I need
I need to be able to use sparse Hessians in R2N.

For example in R2N, the subproblem is constructed by explicitely calling
https://github.com/JuliaSmoothOptimizers/RegularizedOptimization.jl/blob/0f247027945e6fb083fecfd8159bba4deeeacf36/src/R2N.jl#L72.
This means that the subsolver never has access to an other format for `Bk`.

## Why do I need it
In my special case where R2N is used to solve an exact penalty problem, i have an explicit characterization of the solution that uses the inverse of `Bk`, therefore i would like to be able to factorize it.

## How do I plan to do it

I first tried by adding a keyword argument. See this discussion: https://github.com/JuliaSmoothOptimizers/RegularizedOptimization.jl/pull/313#discussion_r2920536406

In place, I proceed by adding an abstraction layer here: https://github.com/JuliaSmoothOptimizers/RegularizedProblems.jl/pull/113.