Add interface for custom scaling

[frapac]

This PR exposes the scaling of the problem to the user with two functions:

• scale_objective
• scale_constraints!

The idea is to dispatch on the AbstractNLPModel being employed. By doing so, the user can overload the two functions to chose a different scaling. For instance, the user can scale the objective by a factor 1e-3with:

function scale_objective(nlp::MySpecialModel, grad::AbstractVector; max_gradient=1e-8)
     return 1e-3
end

This code was already employed to scale the reduced model in the paper.

[codecov-commenter]

Codecov Report

Merging #152 (d4951bb) into master (b582ef3) will decrease coverage by 6.68%.
The diff coverage is 100.00%.

@@            Coverage Diff             @@
##           master     #152      +/-   ##
==========================================
- Coverage   81.22%   74.53%   -6.69%     
==========================================
  Files          31       32       +1     
  Lines        3392     3385       -7     
==========================================
- Hits         2755     2523     -232     
- Misses        637      862     +225     

Impacted Files	Coverage Δ
src/MadNLP.jl	62.50% <ø> (ø)
lib/MadNLPGPU/src/kernels.jl	65.00% <100.00%> (+65.00%)	⬆️
src/interiorpointsolver.jl	95.52% <100.00%> (-0.05%)	⬇️
src/scaling.jl	100.00% <100.00%> (ø)
lib/MadNLPGraph/src/plasmo_interface.jl	0.00% <0.00%> (-93.31%)	⬇️
lib/MadNLPGraph/src/schur.jl	0.00% <0.00%> (-80.24%)	⬇️
lib/MadNLPGraph/src/graphtools.jl	0.00% <0.00%> (-66.67%)	⬇️
lib/MadNLPGraph/src/schwarz.jl	0.00% <0.00%> (-63.77%)	⬇️
src/matrixtools.jl	83.60% <0.00%> (-16.40%)	⬇️
src/utils.jl	47.61% <0.00%> (-10.72%)	⬇️
... and 9 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ec2aa2f...d4951bb. Read the comment docs.

[sshin23]

Great idea @frapac!