[Plugin] SofaCHOLMOD: Add plugin with a supernodal Cholesky direct solver (EigenCholmodSupernodalLLT)

[fredroy]

Add a new optional plugin SofaCHOLMOD providing a direct linear solver, EigenCholmodSupernodalLLT, based on the supernodal sparse Cholesky factorization of CHOLMOD (SuiteSparse), wrapped via Eigen's CholmodSupport.

The dependency on SuiteSparse/CHOLMOD is fully isolated in this plugin so it stays out of the core solvers.

On a ~15k-DOF FEM beam: ~8x speedup over SparseLDLSolver on macOS

Note1: BLAS (important for performance)
Since the factorization delegates to dense BLAS3, performance is dictated by the BLAS backing CHOLMOD, not by SOFA:

• On macOS, SuiteSparse links Apple Accelerate → speedup is always present.
• On Linux, the backend depends on the install BLAS. With the system default one, BLAS the gain mostly disappears; with OpenBLAS/MKL you get the full speedup. See the plugin README for how to check/fix this.
• Multithreaded BLAS tends to oversubscribe on medium systems, so a small thread count is often fastest. A scene-settable numThreads Data is provided to control this at runtime.

Note2: CHOLMOD is double-precision only → enforced at configure time (SOFA_FLOATING_POINT_TYPE=double).

Note3: NOT TESTED ON WINDOWS (YET)
Windows: tested with suitesparse installed through conda (conda install -c conda-forge suitesparse) and using the include and libs there (with miniconda: CHOLDMOD_INCLUDE_DIR=<miniconda_install>/Library/include/suitesparse and CHOLMOD=<miniconda_install>/Library/lib/cholmod.lib)
On a old laptop (i7 11800H), the test in the PR shows:
100 iterations done in 66.1051 s ( 1.51274 FPS). for SparseLDL
100 iterations done in 15.2286 s ( 6.56658 FPS) for EigenCholmodSupernodalLLT

--
EDIT:add implementation of addJMInvJtLocal by Claude:

addJMInvJtLocal accelerates compliance matrix computation by exploiting CHOLMOD’s supernodal factorization,
reducing JA^{-1}J^T to a single forward triangular solve followed by a symmetric rank-k update (Z^T Z). It accesses
the raw CHOLMOD factor through a dedicated proxy and leverages optimized BLAS (dsyrk), providing significant
speedups for medium-to-large systems while small systems remain limited by setup overhead.

on a M3 max (macOS)
TorusFall.scn:
SparseLDL: 5000 iterations done in 14.3029 s ( 349.58 FPS)
EigenCholmodSupernodalLLT: 5000 iterations done in 12.8377 s ( 389.479 FPS)

---

By submitting this pull request, I acknowledge that
I have read, understand, and agree SOFA Developer Certificate of Origin (DCO).

---

Reviewers will merge this pull-request only if

• it builds with SUCCESS for all platforms on the CI.
• it does not generate new warnings.
• it does not generate new unit test failures.
• it does not generate new scene test failures.
• it does not break API compatibility.
• it is more than 1 week old (or has fast-merge label).

[hugtalbot]

Further to SOFA dev