verification: activate the two shell benchmark cases (#106)

The Scordelis-Lo roof and the pinched cylinder had their published
references cited but no recipe, because mesh.block's 2D form is planar
and cannot build a curved patch. Both are now active: the mesh is
written node by node (mesh.node.add / mesh.element.add, the route
docs/verification.md allows), the same cylinder patch the solver's own
benchmarks.rs builds in code, with the quarter/eighth symmetry planes,
the rigid diaphragm and the published load.

- scordelis-lo-roof-shell: 16x16 flat quad4 (MITC4) shells, R = 25,
  t = 0.25, self-weight 90 per area; u_z at the free-edge mid-span
  -0.3024 within 2% (ours -0.29901009815961804, 1.12% low).
- pinched-cylinder: 32x32 flat quad4 shells, R = 300, t = 3, a quarter
  of the unit pinch; u under the load -1.8248e-5 within 2% (ours
  -1.803845342860719e-05, 1.15% low).

Each README carries the generator (so the recipe is reproducible), the
refinement study the tolerance reason cites, and how the reference was
cited before. pinched-hemisphere and mh-twisted-beam now say what they
actually wait for: a doubly curved or helical shell mesh written node
by node, not missing shell elements.

tests/oracle/mh-twisted-beam/expected.json

@@ -3,7 +3,7 @@
  "kind": "benchmark",
  "source": "MacNeal and Harder, 'A proposed standard set of problems to test finite element accuracy', Finite Elements in Analysis and Design 1 (1985) 3–20, Fig. 6 (p. 8), Table 3 (p. 10)",
  "status": "pending",
- "requires": "shell elements (M8) or mapped blocks with helical edges",
+ "requires": "a helical shell mesh: mesh.block's 2D form is planar and takes z = 0, so the twisted strip must be written node by node (as scordelis-lo-roof-shell and pinched-cylinder are), or mapped blocks must gain helical edges",
  "quantities": [
   {
    "id": "in_plane",