feat(plot): holding tabs for cutting mode

[BernardJen]

Closes #63.

Why

A closed contour is severed the moment its loop closes — and the overcut added in #58 exists to make sure of it. That's right for the last piece and wrong for every piece before it: a freed piece can shift or lift under the drag knife while later contours are still being cut, pieces come loose during the job rather than at weeding time, and there's no way to cut a part that should stay put. Overcut and tabs are opposite controls, and both are wanted, per job.

No new idea downstream

The G-code writer already lifts the tool between strokes, so a tabbed contour is simply several strokes and the bridges come out for free. This is a polylines-in, polylines-out transform, and nothing after it — ordering, framing, the time estimate — has to know tabs exist.

Order is the whole design

compensate → overcut → tabs, with "was this closed?" decided once up front and carried through. Tabs have to come last: after them there's no closed contour left for the overcut to recognise, and it would silently do nothing. For the same reason tabs are kept out of the stretch the overcut re-traces — otherwise the overcut cuts through the first bridge and both features look broken at once. (That's the interaction the issue flagged.)

The issue's other worry — that orderPolylines may reverse a stroke and move evenly-distributed tabs — is already resolved: cutting disables reversal (#58).

Details that matter

• Bridges are nudged off sharp corners, where a bridge tears instead of snapping.
• The count is capped by what the contour can afford rather than honoured blindly; fifty tabs on a short contour is a dotted line.
• No degenerate fragments. orderPolylines drops anything shorter than two points, so a collapsed slice would vanish silently and leave an unexplained gap in the cut — they're discarded explicitly instead.
• Off by default, and with tabs off prepareForCut produces exactly what it did before (pinned by a test). A piece that unexpectedly stays attached is as surprising as one that doesn't.
• The canvas draws the prepared cutting path over faded artwork: grey is the drawing, red is what the knife does. Where a piece stays attached is exactly what has to be checked before committing a sheet of vinyl.

Verification

mise run ci green — 385 tests, coverage 66.5 → 67.4% statements. 13 new tests, including no-degenerate-fragments across many count/width combinations, bridges staying off corners, and bridges staying clear of the overcut.

Driven in a browser: 4 tabs of 2 mm on a rectangle produce 5 cut spans in the emitted G-code, with gaps measuring exactly 2.00 mm, evenly spread, each ≥130 mm from a corner.

Not yet verified

⚙ Hardware — a real sticker has to confirm the piece stays attached during the job and snaps out cleanly by hand.

🤖 Generated with Claude Code