PartitionedNLPModel relying on PartitionedVectors

[paraynaud]

@dpo great news, I defined the first partitioned quasi-Newton NLPModels relying fully on PartititonedVectors!
Moreover, trunk() works (from my TrunkSolver constructor).

Locally, the following script is running:

using ADNLPModels, NLPModels
using OptimizationProblems, OptimizationProblems.PureJuMP, OptimizationProblems.ADNLPProblems
using JSOSolvers
using PartiallySeparableNLPModels

adnlp = ADNLPProblems.arwhead(; n)
pqnnlp = PVQNPModel(adnlp)
ps_ges = trunk(pqnnlp)
sol_ps = ps_ges.solution
NLPModels.obj(adnlp, sol_ad) #1.7053025658242404e-13

For now I defined only one PQNNLPModel, I still have to define the others PQNNLPModels and then I will clean up the repository.
Once this stuff is stable, I will release the v0.1.0 of PartitionedVectors.jl and I will register it.
PartitionedStructures.jl will have a new release which will be propagate up to this repository.

[dpo]

@paraynaud Great news! Congratulations! I'm looking forward to seeing it in action!

[paraynaud]

Require new releases of:

• Krylov.jl including this PR;
• NLPModels.jl including this PR;
  until then, the tests will not pass (if I don't make it load the master branch).

[paraynaud]

After an error preventing the build of ExpressionTreeForge.jl due to RecipesBase.jl.
The error resolved itself, I think the PR is ready to go.

[amontoison]

Paul, il y a trop de modifications. Je te fais confiance et j'approuve la PR histoire de ne pas te bloquer.

[paraynaud]

Paul, il y a trop de modifications. Je te fais confiance et j'approuve la PR histoire de ne pas te bloquer.

Merci, les prochaines PR seront plus légères, promis!