[MasoudMiM]

Phase 2 item 3 (the last of the three feature-depth items): detect and rebuild full-revolution ring bands as editable PartDesign history.

What

A rounded ring band is a torus: a tube of revolution standing on a face. A band growing out of the face it sits on is a donut blend the fillet rule cannot claim (fillets are partial revolutions; a ring band is a full one) → torus_boss. A band cut into a face is a ring groove → torus_recess. PartDesign has no torus primitive, so the executor revolve-builds the band: a circular-arc profile sketched on a plane through the major axis, spun 360° as a PartDesign::Revolution (boss) or PartDesign::Groove (seat).

The core had no torus primitive at all, so a ring band was either left unassigned or absorbed by a straight-walled terrace pad (the right footprint at the wrong profile). This adds the full pipeline:

• core/primitives.py: new Torus primitive (complexity 3, 7 dof).
• core/fitting.py: new fit_torus fitter (least-variance axis — a ring is thin along its major axis, opposite a cylinder; circle in the (rho, h) cross-section; nonlinear refinement), registered in fit_best.
• core/emission.py: _torus_patch (poloidal v = 0 at tube top, full-u from the toroidal angle).
• core/features.py: torus_boss / torus_recess rules (rim annulus = two concentric circles where the tube is cut by the face; additive from the rim face's mean normal; axis out of the face for a boss, into the material for a seat).
• core/history.py: TorusOp plan op + BuildPlan.tori.
• core/solidify.py + tests/test_adversarial.py: headless torus-band rebuild for round-trip verification.
• build.py: the revolve/groove executor branch (arcs built through 3 points so >180° profiles close).
• scripts/freecad_smoke_test.py: ring-boss + ring-groove cases.

Impostor gates (the interesting part)

The flexible 7-dof fit grabs impostors a simpler fitter misses:

• Conical — a short frustum's (rho, h) cross-section is arc-like, so the point fit alone can't tell a cone from a torus. A cone's surface normals all make the same angle with its axis, so the fit is rejected when the 95th-percentile normal angle stays below 60° (a genuine torus's normals wrap the tube, 0–90°). This is what was crashing the FreeCAD smoke test: a conical pocket's cone + top face merge into one segment (adaptive angle threshold 60°) and the torus fit won, never re-splitting.
• Partial patch — a torus is a full ring (toroidal angle ~360°). A local curved patch, notably a slot's half-cylinder rounded end (where the cylinder fit fails on the partial circumference), covers only a narrow arc and is rejected when the toroidal-angle coverage is below 300°.

Scope

Rounded ring bosses and ring seats. Standalone donuts (no rim / no planar base) and partial-revolution bands are out of scope.

Verification

• Headless suite: 580 passed, 1 xfailed (was 551 before the torus tests; includes 29 new test_torus.py tests — detection, planning, round-trip, and negatives incl. the slot-half-cylinder impostor).
• FreeCAD 1.1.4 smoke test: 119 checks green, 0 failed (ring-boss and ring-groove volume-exact).