Post-audit reliability hardening — wave 1 (critical + high)

[LeadcodeDev]

Wave 1 of the post-audit chantier — the 2 critical and 8 high findings. Closes the
two silent-failure paths, makes the geometry validator trustworthy, and realigns the
authoring documentation with the schema the code actually accepts.

Parent: #102 · Sub-issues: #103 #104 #105 #106 #107

One commit per workstream; each is independently reviewable.

What was broken

	Before	After
--fix on an unwrappable overflow	wrote style.wrap, a field CssStyle rejects → component dropped from the render, validation then reported clean	removes style.white-space, falls back to wrapping
#fff / white / rgb(255,255,255)	resolved to black, silently, on dark backgrounds	render correctly; unresolvable colours warn and render magenta
grid with three 1fr tracks	three stacked full-width rows; tracks never reached taffy	three columns, pixel-identical to the flex-row equivalent
rustmotion schema	Scene.children typed "items": true — no component described	$ref to a 57-variant oneOf; 38 → 178 definitions
Content overflowing a fixed-height card	never reported (both boxes stay inside the frame)	new content_overflows_box violation
Frame-bleeding typography under overflow: hidden	rejected as a viewport overflow	validates clean — clipped bleed is a technique, not a defect
--strict-anim	discarded t=0, the frame of maximum displacement	samples proportionally to duration, no opacity short-circuit
--fix with a malformed sibling	patched an unrelated component	raw JSON index preserved
Validation text measurement	used the fallback face while the render used the declared one	fonts loaded before measuring

Verification

cargo test --workspace → 534 passed, 0 failed. cargo build --release clean.
All 6 examples/*.json still validate.

Beyond the per-workstream tests, an independent acceptance harness re-derived every
audit reproduction from scratch — deliberately reusing none of the fixtures the
implementing agents wrote — including pixel measurement of rendered frames:

baseline (pre-fix binary):   3/11 green
after wave 1:               11/11 green

examples/mega-showcase.json changes visually: its 2×2 feature card now renders as a
real grid instead of four stacked rows. Verified by eye — this is the H1 fix landing,
not a regression.

Deliberately deferred to wave 2

• A validation-time error for unresolvable colours. This PR makes the failure loud
  at render (magenta + stderr) but does not yet fail validate.
• schema::GridTrack, the last stale shadow type. Deleting it means repointing a
  re-export in traits/container.rs, outside this wave's file partition.
• schema::style::Spacing and GridPlacement, dead once LayerStyle is gone.

Merge-order note

crates/rustmotion-core/src/engine/renderer/fonts.rs has a deeper bug than the one
fixed here: register_font_file discards the Typeface returned by
FontMgr::new_from_data, so declared families are never resolvable by name. It was
rediscovered independently during this work — and it is already fixed by 208e8d1
on docs/1600-brutalist-style, which is not yet on main. No follow-up issue filed
for that reason.

Consequence: the font fix in this PR (measuring with the declared face rather than the
fallback) only takes full effect once that branch lands. When both are in, geometry
results may legitimately shift for scenarios using custom fonts.