[Small Feature] OSQP Support in Solver Interface

[danston]

This PR adds the OSQP solver support in CGAL. This solver is required in order to efficiently compute the convex Quadratic Programming (QP) problem, arising in the context of the new CGAL package called Shape Regularization (see #5742). This PR also better structures the Solver_interface package by splitting all solvers into several common categories. An additional example is added, too.

Tasks:

• Check memory leaking.
• Remove tmp modif in the docs.

Release Management

• Affected package(s): Solver_interface
• Issue(s) solved (if any): no issues
• Feature/Small Feature (if any): Small Feature pre-approved on June 28, 2021
• Link to compiled documentation (obligatory for small feature): docs
• License and copyright ownership: no change

[afabri]

• Instead of #if defined(CGAL_USE_OSQP) deal with it in the CMakeLists.txt
• The brief documentation should start lower case
• The sentence "It is also possible to derive new models from other quadratic programming libraries such as OOQP for example." is not really of interest.
• "Note that you should .. " Use \note Also "should" makes no sense in a spec. Do you mean "must"? Or is it correct, but a waste of time?
• "all existing/previous entries". Remove existing and previous and maybe even the entire sentence.
• So this solver supports any FT ?
• Does one use just the C OSQP, or the C++ wrapper?

[danston]

@afabri Thank you for the review.

• Instead of #if defined(CGAL_USE_OSQP) deal with it in the CMakeLists.txt - And what about other examples? The mixed-integer example in this package has been written this way that is it is dealing with libraries inside the cpp. I just followed that style. Should I change them as well?
  -- I fixed it both for the osqp and mixed_integer examples.

• So this solver supports any FT ? - As input yes, internally everything is converted to double because the OSQP requires c_float types.

• Does one use just the C OSQP, or the C++ wrapper? - It is the C version.

[afabri]

It is hard to see which review items are done and which not. Maybe copy and mark done as we do on the wiki.
Concerning the FT template parameter I am not sure that it makes sense if the package only works with double. @sloriot @lrineau your opinon?
@MaelRL had problems with underconstrained systems. Is that now fixed? Can it be underconstrained? If not then this merits maybe a \warning.
Is a c_float a float or a double?

[danston]

Instead of #if defined(CGAL_USE_OSQP) deal with it in the CMakeLists.txt

• Done. I fixed it both for the osqp and mixed_integer examples.

The brief documentation should start lower case

• Done. I fixed all brief.

The sentence "It is also possible to derive new models from other quadratic programming libraries such as OOQP for example." is not really of interest.

• Done. I removed it.

"Note that you should .. " Use \note Also "should" makes no sense in a spec. Do you mean "must"? Or is it correct, but a waste of time?

• Done. I rephrased that.

"all existing/previous entries". Remove existing and previous and maybe even the entire sentence.

• Done. I removed that.

So this solver supports any FT ?

• The wrapper yes, but internally it is C float.

Does one use just the C OSQP, or the C++ wrapper?

• It is the C version.

[danston]

Errors + warnings on Windows: https://cgal.geometryfactory.com/CGAL/testsuite/CGAL-5.4-Ic-5/Solver_interface_Examples/TestReport_Sosno_MSVC-2022-Preview-Release.gz

[maxGimeno]

tested in https://cgal.geometryfactory.com/CGAL/testsuite/results-5.4-Ic-15.shtml