t27f: reel_beats.t27 -- every reel an ordered open, main and close scene sequence, distinct across the course (R7 of #10560) #10584

Sub-issue of epic #10560 (content factory). R7.

Boundary

• New spec specs/t27f/reel_beats.t27, its seal, and its t27b ledger row. It reads specs/t27f/reel_factory.t27 (t27f: reel_factory.t27 -- content features to a unique scene plan per item #10561, PR t27f: reel_factory.t27 -- content features to a unique scene plan per item (Closes #10561) #10566) and does not change it.
• Not in scope: renderers (R3, t27f factory R3: a renderer per scene archetype and the still anomaly gate (SSIM) #10571), lesson_reel reading the plan (R4), beat timing in frames.

User Scenarios

1. The owner said every lesson reel looks almost the same. reel_factory.t27 gives each lesson one scene archetype, capped at 4 uses, so up to four reels of a course share their only scene and differ only in accent, motion and layout. Its own NOT CLAIMED block says the signature-uniqueness test is weak by itself.
2. Competitors (notes: content-factory-competitors.md): Motion Canvas / Revideo let the data choose the scene LIST per project; slot-filler tools (Canva, Creatomate, Shotstack templates) fill one fixed structure, which is the sameness the owner sees.
3. With R7, each reel is an ordered sequence of three beats: open, main, close. Main is the factory's scene. Open is the item's next-best content scene. Close is a generic scene (code, title, task or map) chosen so that the sequences stay apart.

Requirements

• R7.1 main(t) = arche(t); open is the non-generic archetype the item has content for, other than main, with the best term-frequency score; a generic scene only when the item has no second construct.
• R7.2 close is generic, differs from open and main, and rotates so that neighbours do not close alike.
• R7.3 Tests over all 27 lessons of course 0: three distinct scenes per reel; all 27 ordered sequences distinct; no two reels share both main and open; two neighbours share no beat in the same position; open shows content the item has.
• R7.4 0 vacuous tests (t27c test-report), seal saved and verified, t27b ledger row measured with a t27b that includes t27b: std.debug.print lowers by a plan in t27 -- zig_open_slice passes, reel_particles_mux loses its std.* blocker (Closes #10449) #10455.

Success Criteria

• t27c test-report specs/t27f/reel_beats.t27 passes every test with asserts executed, on the Railway lab.
• A printed sequence table for course 0 in the PR body.
• Seal --verify MATCH; ledger cap max_not_pass does not rise.

Requirements

• FR-001: The specification MUST define main(t) = arche(t) for each reel t.
• FR-002: The specification MUST define open as the non-generic archetype the item has content for, other than main, with the best term-frequency score; a generic scene only when the item has no second construct.
• FR-003: The specification MUST define close as a generic scene that differs from open and main, and rotates so that neighbours do not close alike.
• FR-004: Tests over all 27 lessons of course 0 MUST verify: three distinct scenes per reel; all 27 ordered sequences distinct; no two reels share both main and open; two neighbours share no beat in the same position; open shows content the item has.
• FR-005: The specification MUST produce 0 vacuous tests under t27c test-report, save its seal, verify the seal, and produce a t27b ledger row measured with a t27b that includes t27b: std.debug.print lowers by a plan in t27 -- zig_open_slice passes, reel_particles_mux loses its std.* blocker (Closes #10449) #10455.