Outcrop on curved boundaries: the wall descriptor is (axis, value) everywhere

[lmoresi]

The outcrop machinery (3-D: #525/#526; 2-D: PR #552) is restricted to axis-aligned box walls, and the restriction is ours, not gmsh's. OCC's intersect clips against any tool solid — a disk, cylinder, or revolved shell works exactly like addBox — and gmsh can mesh cap patches on curved OCC surfaces or in a local chart. What blocks a curved outcrop (a fault daylighting on an annulus top, a spherical surface) is that the wall is represented as the pair (axis, value) threaded through everything:

• band identification is coordinate-equality against the plane/line value (_split_skin_band);
• the post-mesh snap is per-axis (_occ_assembly_2d / _occ_assembly_3d);
• the carve's open-wall tests are X[:, axis] == value (victim protection, ring checks, the 2-D splice, the 3-D bowl);
• the 3-D cap is meshed flat in the wall plane's uv coordinates.

Generalising means replacing the pair with a small wall descriptor — implicit function for membership, projection for the snap, chart for the cap — a bounded refactor: one concept, several touch points, each already isolated behind a named test in the 2-D path.

The annulus fault studies are the first customer. Corner/box-edge outcrop (a zone meeting two walls) is a separate, genuinely harder problem and stays refused regardless.

Underworld development team with AI support from Claude Code

[lmoresi]

Progress in #582: the assembly clip now cuts against the mesh's own boundary rebuilt as OCC geometry (2-D and 3-D), the outcrop band is identified as the trace on any boundary and labelled <label>_trace, and the 2-D carve/sew is fully general — a radial ribbon out of the annulus's outer boundary embeds with area conserved to 1e-12 and the P2 oracle exact through the zone, at np=1/2/4 with partition-independent counts.

Remaining for this issue: the general 3-D cap. A zone outcropping on a non-axis-aligned 3-D boundary is currently refused with the reason (TODO(DESIGN) at _trace_wall_code). The construction we intend: re-triangulate the bowl-minus-band collar per coplanar boundary region, in each region's own plane so the cap stays exactly on the faceted surface, with 1-D overlays where the band outline crosses creases. That is a spike-validated piece of work of its own.