fix: connector_detached now skips fully dash-hidden connector shafts

[miga-heygen]

Summary

connector_detached (the layout-audit finding for SVG connector paths whose rendered endpoints don't attach to any anchor) evaluated a path's geometry unconditionally — it had no gate for stroke-dashoffset/stroke-dasharray draw-on visibility. A connector fully hidden behind an un-advanced "draw the line in" dash animation still fired the finding as if it were rendered and visible. The sibling check, connector_orphan, already skipped fully dash-hidden shafts via shaftDashHidden(path).

Two things changed on top of simply adding that gate:

One owner for connector enumeration. connector_detached and connector_orphan each walked svg path with their own copy of the candidate/gate logic (visible svg, skip defs/marker/clipPath/mask/symbol/pattern, connector intent, endpoint resolution, dash gate) and the copies had drifted. A single connectorShafts(root) generator now owns that walk and yields { svg, path, user, rendered, chord, painted }; both findings consume it. connector_orphan still requires painted (display/visibility/opacity); connector_detached still does not gate on it — that is unchanged behaviour and documented at the owner. The path length is queried once per shaft (pathUserEndpoints now returns total) instead of twice, and the attach-threshold formula has one owner.

shaftDashHidden reads the whole pattern. It previously read only the first stroke-dasharray entry, so a visible dotted pattern like 0 4 — or any pattern with a long first dash — was treated as hidden and suppressed both findings. It now resolves the full computed list (comma/space separated, units, percentages against the SVG viewport's normalised diagonal, odd lists repeated; none, all-zero and negative lists render solid) and treats the shaft as hidden only when the visible window [dashoffset, dashoffset + length] sits inside a single gap, allowing up to 10% of the length to overlap a neighbouring dash (the tail of a nearly finished tween — same boundary the old >= 0.9 * length rule had). A zero-length dash paints only as a round/square line cap; under the default butt it renders nothing, which also covers the 0px, 999999px state a finished draw-off tween leaves behind.

Test plan

• Regression test: a connector fully hidden via strokeDasharray/strokeDashoffset no longer fires connector_detached, on the same fixture as the adjacent positive control (shared foreignFrameDom/foreignFrameRects/foreignFrameStyles, mirroring this file's orphanDom convention).
• Dash-pattern table (path length mocked to 100): 0 4 with butt cap → hidden; 0 4 round and 0, 4 square → painted; 0px, 999999px @ -99.999px → hidden; 0 400 round @ 0 → painted (dot inside the window), @ 1 → hidden; 4 0, 0, none → painted; 100px @ 100px and 100 @ -100 → hidden; 50 100 @ 50 and @ 40 (exactly 10% painted) → hidden, @ 30 → painted.
• Existing connector_orphan tests (including its dash-hidden case) pass unchanged against the shared enumerator.
• layout-audit.browser.test.ts does not collect under this sandbox's vitest (No such built-in module: node: — pre-existing, also on an unmodified checkout). Run instead under bun test with a happy-dom preload: 126 pass, 9 fail, where the 9 are contrast-audit cases that fail identically before this change. bunx tsc --noEmit -p packages/cli, oxlint, oxfmt --check clean; pre-commit lint/format/fallow/typecheck hooks pass.
• Not verified in a real browser: % dash lengths (resolved against viewBox, else the SVG's layout box) and calc() values (parse to NaN → treated as solid, i.e. the safe direction).

[miga-heygen]

Addressed in 58aefbb.