V2.7.2mod: dynamic loading for HSL + improved build

[jgillis]

Hi Charlie!

Finally found some time to get Uno into casadi..

I did some minor modifications to get it more in line with expected behavior from ipopt.
I'm happy to just roll with casadi-patches against your official releases, but you may also chose to merge some of this.

[cvanaret]

Hi Joris,
Thanks for the contribution! Great to see dynamic loading for the HSL solvers.
I will definitely add a reverse CI workflow to run Uno via CasADi, so that we know when changes break.

@amontoison could you have a look when you have a moment?

[amontoison]

Thanks @jgillis for the modifications and working on an interface for CasADi!

A few comments:

• I am a big fan of the dynamic loader for HSL.
  For Julia, it is easy to do hot-swapping, so I precompile a dummy version of libHSL with the required symbols such that the
  linker is happy, and at runtime we call the routine LIBHSL_ISFUNCTIONAL() to check what is available. It works quite well when the
  user has access to the installation folder, because they can replace libhsl.$dlext on their own.
  So it's quite nice when the user downloads the precompiled artifacts from GitHub. Julia does that automatically too with HSL_jll.jl.
  But your approach is the best one if Uno is installed through some other way like PyPI, Homebrew, a Linux distribution, etc. We can't easily hot-swap the library.

• For METIS, why are you not using -DAUXILIARY_LIBRARIES?
  It is dedicated to dependencies of dependencies for static builds.

• For the header of SPRAL / SSIDS, Charlie made a modification to add a CMake variable but it never landed in main. Are these modifications enough for you?

[cvanaret]

The CMake variable SPRAL_INCLUDE_DIRis now available (see #779). I have updated the PR and simplified the CMake a bit.
As Alexis wrote, we can handle METIS via AUXILIARY_LIBRARIES. I'll do a proper read of the dynamic loader in the next few days.

[cvanaret]

Hi @jgillis, I have merged a separate PR (#792) to optionally build with OpenMP, and updated your branch.
I'm working on a paper review at the moment (due Thursday), so I'll have a look at the dynamic loader at the end of the week.