t27f: reel_factory.t27 -- content features to a unique scene plan per item #10561

Part of #10560 (R1 + R2).

The 27 lesson reels share one backdrop: token_floor's physics world. This issue adds the factory law that gives each item its own scene, derived from its own content and checked for uniqueness by tests.

Boundary

• In: a new specs/t27f/reel_factory.t27, its seal, its t27b ledger row, and a row in specs/t27f/frontend.t27's header that points at it.
• Out: the renderers of the archetypes (R3) and rewiring lesson_reel.t27 (R4). Those are later issues.

User Scenarios

1. The lab prints the plan of course 0 with t27c run specs/t27f/reel_factory.t27. That gives one row per lesson: its archetype, accent, motion, layout and features.
2. A reviewer reads the tests and sees that uniqueness is checked over all 27 lessons, not asserted.

Requirements

• Features are counted from each lesson's code in reel_topics.t27 (keywords at word boundaries). They are never typed by hand per lesson.
• The archetype is the best-scoring one that the lesson has content for, is not the previous lesson's archetype, and is not over its cap.
• The accent comes from the course module. The motion is t mod 4 and the layout is t mod 3.
• The output formats are tall and wide reels, a square post, a 4:5 post, a cover, a carousel, an X card and an og image. Each has its size and its carousel slide count.
• The anomaly gate for R3 is a SSIM ceiling between any two rendered stills.

Success Criteria

• t27c test-report on the lab: every test passes, 0 vacuous.
• Witness tests tie the keyword table to real lessons: the test-block lesson has tests, the loops lesson has loops, the structs lesson has data, and so on.
• t27c seal --verify matches. The t27b row reads pass.