recipe 03 revolves a wire, so the flange is a shell rather than a solid

[gsdali]

Found while diagnosing the Route B failure in #86.

The defect

recipes/03-pipe-flange/main.swift:45 builds the flange by revolving a wire:

let section = Wire.polygon([...])!
var flange = Shape.revolve(profile: section, axisOrigin: .zero,
                           axisDirection: SIMD3(0, 1, 0), angle: 2 * .pi)!

Shape.revolve(profile:) revolves a curve, so it correctly returns a shell, not a solid. This
is standard BRepPrimAPI_MakeRevol behaviour and not an OCCTSwift bug. Rebuilt and measured:

recipe03 flange: type=shell  valid=true  vol=228707.95  faces=6

The recipe then chamfers it and cuts a bolt circle, treating it as a solid throughout. Its README
describes the output as a solid body.

Why CI does not catch it

Scripts/recipe-check.sh asserts the emitted volume is greater than zero. A closed shell
returns a correct positive volume from volumeInertia, so the check passes. Nothing currently
asserts the result is a solid.

The committed recipes/03-pipe-flange/output.brep reports shapeType: compound with volume
241242.9, which looks healthy by every metric the recipe suite measures.

Why it matters

A shell is a surface model. It will not fillet reliably, exports to STEP as surfaces rather than a
solid body, and misbehaves as a boolean operand. In #86 the same pattern produced disconnected
surface patches once real cuts were attempted against it, with Shape.volume returning nil and
the written BREP giving a negative volume. Recipe 03 gets away with it because its bolt-circle cut
happens to leave a closed shell, but it is the same latent error and it is being shipped as a
reference example.

Fix

Revolve a face:

guard let section = Wire.polygon([...]),
      let face = Shape.face(from: section),
      let flange = face.revolved(axisOrigin: .zero, axisDirection: SIMD3(0, 1, 0)) else { ... }

Verified on the #86 blank: Shape.face(from: wire).revolved(...) yields type=solid with the
same volume, and downstream booleans then behave correctly.

Suggested follow-up

Two things worth doing alongside the fix:

1. Audit the other recipes for the same pattern. Any Shape.revolve(profile:) or
   Shape.extrude(profile:) whose result is treated as a solid should be checked.
2. Strengthen recipe-check.sh to assert the emitted body is a solid, not merely that its
   volume is positive. That is the assertion which would have caught this, and it is the same
   class of gap as a check that cannot fail.

[gsdali]

Scoping: two recipes affected, not three, and extrude is not implicated

Measured solids / shells sub-shape counts on every committed recipe output, plus the two
construction primitives directly.

extrude is fine, revolve is not

Shape.extrude(profile: wire, ...)        : type=solid  solids=1  vol=1000.0
Shape.face(from: wire).extruded(by:)     : type=solid  solids=1  vol=1000.0

Shape.revolve(profile: wire, ...)        : type=shell  solids=0
Shape.face(from: wire).revolved(...)     : type=solid  solids=1

Shape.extrude(profile:) faces the wire for you; Shape.revolve(profile:) does not. That
asymmetry is the actual trap here, and it is worth raising upstream as an API consistency
question: two neighbouring sweep factories taking a Wire and returning different topological
kinds is surprising, and nothing in the signature hints at it.

So recipes 01 and 04, which use extrude, are not affected. Only the revolve path is.

Audit of committed recipe outputs

recipe	type	solids	shells	verdict
01-mounting-bracket	compound	1	1	fine
02-helical-spring	shell	0	1	affected
03-pipe-flange	compound	0	1	affected
04-spur-gear	compound	1	1	fine
05-lattice-cube	compound	1	1	fine
06-fan-blade	compound	1	1	fine
07-sheet-metal-channel	compound	1	1	fine

shapeType alone is misleading: five of these report compound and are perfectly healthy,
because a compound wrapping one solid is the normal result of circularPatternCut. The
solids count is the assertion that actually separates them.

Recipe 02 is a second instance, reached by a different route: it is a pipe sweep rather than
a revolve, and it emits a bare shell. Worth fixing in the same pass since it fails the same
assertion.

Implication for the recipe-check.sh hardening

Assert solids >= 1 on the emitted body, not shapeType == solid. The latter would reject the
five healthy compounds. That distinction is the whole reason this went unnoticed.