t27f: R4e -- mode 5 cards live inside one scene too (token_floor_mux takes a caller window) #10628

Follow-up to R4d (#10614, PR #10627), epic #10560.

In LR_MODE=5 the captions, tags and address now stay inside one scene. The cards still sit over the beat crossfades, because token_floor_mux prints their enable= on its own frames and lesson_reel cannot reach it.

Boundary

• specs/t27f/token_floor_mux.t27 (the card printer, print_fade or its caller) and specs/t27f/lesson_reel.t27, plus their seals.
• Out of scope: the beat clock itself, modes 3 and 4 (their output must stay byte for byte the same), and the scene clips (R3).

User Scenarios

• A viewer watches a mode 5 reel. No card fades in over one scene and out over the next. Each card appears after a crossfade ends and leaves before the next one starts.
• An author on mode 3 or mode 4 gets the same graph as before.

Requirements

• The card printer takes a window supplied by the caller; with no window it prints what it prints today.
• Mode 5 passes beat_piece_* windows to it, the same function the captions use, so cards obey the same drop-not-squeeze rule.
• Every other caller of token_floor_mux is unchanged.

Success Criteria

• On the Railway lab, t27c test-report passes with 0 vacuous tests for both specs, and both seals are saved and verified.
• The R4d grid test is extended to cards: across every lesson, both languages and every voice the clock accepts, no enabled card frame is on a crossfade. A control shows that the current cards are on one.
• The cmp regression for modes 3 and 4 against the PR t27f: R4d -- mode 5 overlays live inside one scene, none on a beat crossfade #10627 binary finds 0 differences, and so does every other graph token_floor_mux prints.
• ffmpeg renders mode 5 from solid-colour clips. A band over the cards shows no text on any crossfade frame.

Success Criteria

• t27c test-report specs/t27f/token_floor_mux.t27 specs/t27f/lesson_reel.t27 exits 0 with 0 vacuous tests and both seals saved and verified
• t27c test specs/t27f/lesson_reel.t27 --grid cards exits 0 with 0 enabled card frames on crossfades across every lesson, language, and voice the clock accepts
• cmp mode3.bin mode3_ref.bin && cmp mode4.bin mode4_ref.bin exits 0 (0 differences) for modes 3 and 4 regression against PR t27f: R4d -- mode 5 overlays live inside one scene, none on a beat crossfade #10627 binary, and every other graph token_floor_mux prints matches byte for byte
• ffmpeg -hide_banner -i solid_color_%04d.png -filter_complex "[0:v]drawtext=fontfile=/dev/null:fontcolor=white:fontsize=24:x=10:y=10:text='%{pts}'" -f framemd5 - produces framemd5 where the card band region shows no text on any crossfade frame (0 frames with text on crossfade)