WIP (parked): folding robustness, stabilization, mesh.revolve, boot-seal profile

Parked work for #182/#185 (not for merge as is):
- contact: held active set with primal-dual updates replaces the
  inconsistent ENGAGED_BAND tangent after chattering (moves the
  nl-friction-sliding-blocks fixture by 4.5e-8; not re-blessed)
- nonlinear: opt-in viscous stabilization (step 'stabilization',
  calibrated on the first increment), Abaqus STABILIZE, LS-DYNA refusal
- ordering: graph (level-structure) nested dissection, chosen when it
  fills less: half the fill on the revolved boot, factor time 62 s -> 25 s
- engine: mesh.revolve command (2D profile -> hex8/wedge6, tags kept)
- engine: sample/boot.rs boot-seal profile, axisymmetric and 3D scripts
Scratch still in the code: FC_TRACE/TIMES timing prints, NO_BAND/STAB/
SQUEEZE env hooks in tests, boot_axisymmetric_scratch test.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UhhZMBTviDQxZvwnyP89fo

crates/decks/tests/lsdyna.rs

@@ -774,6 +774,9 @@ fn nonlinear_static_steps() {
         .collect();
     assert_eq!(curve, [vec![0.0, 0.0], vec![0.5, 1.0], vec![1.0, -1.0]], "{deck}");
 
+    m.replace_step(step("Bend", Procedure::NonlinearStatic(NonlinearStatic { stabilization: Some(2e-4), ..controls.clone() }), &["clamp"], &["tip_load"]))
+        .unwrap();
+    unsupported(&m, "stabilization");
     m.replace_step(step("Bend", Procedure::NonlinearStatic(NonlinearStatic { nonlinear_geometry: false, ..controls }), &["clamp"], &["tip_load"]))
         .unwrap();
     unsupported(&m, "a nonlinear step without nonlinear geometry");