feat(rows): the future folds — 7d tail token, hour cells, forecast guard, meter demo

[lroolle]

Why

Two 7d burnouts in a row exposed both halves of this PR. A stale pre-envelope build had poisoned forecast.cache (recent_24h: 343, Thursday at 145%/day of a 100-point pool) and the walk called a 2%-used fresh window dry in 30 hours, red, on every render. And even with sane data, the 7d strip spent 30 columns drawing hollow future cells that all said one thing.

What

• 7d strip folds its future (live row only): history in full, ▮, two kept cells, then ...▯5hx28 — count of 5h slots to the reset, × red when the tail projects dry. Folds only when it hides ≥ 10 cells; the week report still draws all 34.
• 5h strip by the hour: 5 cells, was 10 × 30 min.
• Forecast guard: _profile_walk treats a weekday profile > 100%/day as corrupt (no verdict beats a red siren) and clamps recent_24h to the pool (legitimately larger across a reset, unusable as a within-window rate).
• Pages demo redesigned as the instrument it demos: no fake terminal chrome — a recessed register running the same frame simulation (fold, hour cells, dry-forecast frame), № edge-codes cited by answering-pair anatomy rows, odometer digits, self-served block-glyph subset, truthful narrow-screen compaction.
• Drift killed: 417 tests stated everywhere (README said 383, site 384, llms.txt 379); llms.txt learned the new week-row grammar.

Verify

• npm exec --yes bats -- t/ → 417 pass (2 new fold tests, 2 new guard tests; layout test moved to COLUMNS=88 since the narrower strips merge at 96)
• shellcheck -S error statusline.sh install.sh → clean
• Rendered against a real 5.3 MB usage log; page captured at 390/768/1024/1440 — no page-level horizontal scroll, glyphs pixel-verified

🤖 Generated with Claude Code