fix(validate,info): line bounding boxes, rendered duration, and four doc claims

[LeadcodeDev]

Closes #369. Closes #372 (partly — see below). Closes #370. Refs #388.

#369 — a line's bounding box ignored its endpoints

line, arrow and connector paint at their own x1/y1/x2/y2 inside a canvas already translated to the node's layout origin, but the validator reused bbox_of(layout): right size, wrong origin. So an overflow was reported in the wrong place, and a real one with a negative x1 passed silently.

The canvas-translation behaviour was confirmed empirically before trusting the issue's formula — a line at x=300,y=200,x1=100,y1=100 puts ink at (400,300), i.e. x + x1, so the endpoint offset genuinely adds to the layout origin rather than replacing it.

Red proof:

bbox must be anchored at x1/y1 (960, 100), not at the node's own x/y (0, 0):
  left:  (0.0, 0.0, 1.0, 1180.0)
  right: (958.0, 98.0, 4.0, 1184.0)

a line whose x1 pokes past x=0 must be reported even though its own box (x=0) does not: []

Against the release binary, the issue's JSON now reports [958, 98] -> [962, 1282] instead of [0, 0] -> [1, 1180].

#372 — info re-derived the duration instead of asking the encoder

validate already used build_frame_tasks(...).len() as ground truth. info was the last place summing scene.duration, so it ignored v1 transition overlap and v2 at placement entirely. It now calls the same function render and validate do.

repro	before	after
two 1.0 s scenes, 0.6 s transition (v1)	60 frames	42 frames / 1.4 s
at: 0 and at: 1.0 over 2.0 s scenes (v2)	120 frames	90 frames / 3.0 s

Part 2 of #372 is not fixed here, deliberately. A scene duration landing on a half frame adds one — two 1.05 s scenes render 64 frames, not 63 — because build_frame_tasks rounds each scene independently and the error accumulates. That is in encode/video/tasks.rs, outside this workstream's ownership. info now faithfully reports what render will produce, so the two no longer disagree with each other; both still disagree with what rounding the cumulative timeline once would give. I am keeping #372 open for that half rather than closing it — see the note below.

#370 — four documentation claims, each verified against the binary

The brief was to establish which side was wrong for each, and to report rather than document a bug as intended behaviour. All four turned out to be the doc drifting from working behaviour — no engine bug hiding behind any of them.

Claim	Verdict	Evidence
div defaults to border-radius: 12	doc wrong	corner pixel reads the background colour — sharp. Every decoration painter does .unwrap_or([0.0; 4])
scale_x/scale_y are animatable	doc wrong	unknown animation property 'scale_x': expected one of … scale, scale.x, scale.y; scale.x validates
layout.padding takes an object	doc wrong	invalid type: map, expected f32. SceneLayout.padding is Option<f32>; CssStyle.padding is Option<Edges> — the doc conflated two different fields
commands take a positional path	doc wrong	error: unexpected argument found; --help shows --file <FILE> required, no positional

The scale_x asymmetry is real but pre-existing: translate_x/translate_y are accepted in underscore form and only scale differs. Left alone — nothing is broken, the naming is just inconsistent, and schema/video.rs was out of scope.

Verification

cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace (1594) all clean. Comment count still 0.

Written comment-free, per the codebase-wide rule from #345.