fix(check): time a line the next scene repeats in place as one line - #7704

Context: I've been using HyperFrames every day for the past two weeks to make Korean Shorts and long-form videos, and it's been great to work with. This PR is one of a batch of fixes for things I ran into along the way, so you'll see several PRs from me at once — sorry for the pile-up, and thanks for your patience. Each one is reproduced on current main, independent of the others, and can be reviewed, merged or closed on its own.

What

line_unread now measures a line that a scene cut carries over (the same words, in the same place, in the next sub-composition) as one line, so its hold runs across the cut instead of stopping at it.

Why

The craft pass tracked each DOM element on its own. A film that keeps an alert on screen across two scene files (scene A ends at 2.1 s, scene B starts at 2.1 s with the same <div> text in the same place) showed the alert for 7 s, yet check reported the first copy only:

line_unread | compositions/a0.html #h-a0 > div:nth-of-type(1) > div:nth-of-type(1) |
"BALLISTIC MISSILE THREAT INBOUND TO HAWAII. SEEK IMMEDIAT..." (14 words) is on screen 2.10 s (±0.01 s), under its 3.53 s floor

This came from a real Short whose alert card continues from its cold open into the next scene.

Related work

• No open issue or PR covers it (searched line_unread, sub-composition, craft).
• fix(cli): count line_unread words from the block's rendered text #6456 (block text for word counts) and feat(cli): a line_unread finding carries the frames it measured around where the line was first read #7294 (measured frames) touched the same tracker; this keeps both.

How

• trackBlocks (packages/cli/src/utils/craftAudit.ts): when a block is seen for the first time, and a block in the previous sample showed the same text in an overlapping box, the new block joins that block's track. A track now holds a set of block keys.
• The edge re-measuring probe (refineWindow) and measuredAround check any of the track's keys.
• The same words somewhere else on screen still count as a separate line (the boxes have to overlap).
• Docs: docs/packages/cli.mdx craft table and skills/hyperframes-cli/references/lint-validate-inspect.md (manifest regenerated).

Test plan

• New unit test measures a line the next scene repeats in place as one line fails on main (expected [ { code: 'line_unread', … } ] to deeply equal []) and passes with the fix; a control test keeps the same words placed elsewhere as their own line.
• bunx vitest run src/utils/craftAudit.test.ts: 29 passed.
• Real CLI on a three-scene repro (scene a0 0–2.1 s and n1 2.1–7 s show the same 14-word alert, e1 shows "THE END"), hyperframes check --no-cache --json:
  • main (0.9.43): 1 line_unread finding, the alert at 2.10 s under its 3.53 s floor.
  • this branch: 0 craft findings.
• On the original project (alert carried across a0.html → n1.html), the two carried-over lines (BALLISTIC MISSILE…, THE REAL ALERT · JAN 13, 2018…) no longer warn.
• oxlint, oxfmt, fallow audit --base origin/main, comment ratchet, typecheck (pre-commit).

🤖 Generated with Claude Code