Allow smooth box constrained problems

[MaxenceGollier]

With JuliaSmoothOptimizers/ShiftedProximalOperators.jl#156, it should now be possible to use our solvers for unregularized problems.

Users should be able to call the solvers with regular AbstractNLPModels, for example

nlp = CUTEstModel(...)
solver = R2NSolver(nlp) # Automatically creates a reg_nlp = RegularizedNLPModel(nlp, NullRegularizer())
stats = RegularizedExecutionStats(nlp)
solve!(solver, nlp, stats)

[MaxenceGollier]

I think this is not a good idea because we'd need to allocate a new reg_nlp at the start of the solve!. I implemented the functions solver(nlp; kwargs...) though in #305.

I think having a new RegularizedNLPModel(nlp) constructor that calls RegularizedNLPModel(nlp, NullRegularizer) would be great here.