Beam compression validation, plane strain, SOFA vs FF

[Fimache]

Adds compression validation for the 2D cantilever beam
(fully clamped left edge, traction load on the right edge), plane
strain. No analytical solution is exact over the whole domain
(Saint-Venant boundary layer near the clamp), so validation combines:

1. SOFA vs FreeFEM cross-validation (primary check, full domain)
2. A far-field analytical check (x >= 2H) against the uniform
   uniaxial compression solution

• Elements: P1 triangles (beam2d_tri.msh, L=1, H=0.2)
• BCs: left edge fully clamped (ux=uy=0); traction (-q, 0) applied
  on the right edge via consistent nodal forces (line-integral,
  1/2 per node per boundary segment)
• Plane strain: template="Vec3d" (lambda = E*nu/((1+nu)(1-2*nu))
  on the FreeFEM side)
• Params: q=100, E=1000, nu=0.3

Results

SOFA and FreeFEM agree at machine precision across the entire domain.

Files added

• beam2d_tri.msh — shared mesh (already present from the
  distributed-load / plane stress PRs)
• sofa_beam2d_compression.py — SOFA scene (plane strain)
• freefem_beam2d_compression.edp — FreeFEM weak form (plane strain)
• comparaison_script_2d_compression.py — cross-validation driver
• params_beam2d_compression.json — physical parameters (q, E, nu)

[th-skam]

Also asymmetric u_y

Since all your files have the same name, and they do kind of the same thing, give a bit descriptive names to your directories:
Compression vs compression_cp

[th-skam]

It's ok, you just had gravity on.
I'll remove it in the factorized version.