Stroke control, computed path data, and referenced clip paths

[LeadcodeDev]

Part of #326. Batch 3 — depends on #327 and #328.

Rustmotion has roughly fourteen drawing primitives against about twenty for Canvas and fifteen for SVG, so the count is not the problem. ClipPath::Path and Polygon already exist, mix-blend-mode exists, shape: {path: {data}} accepts arbitrary path data, and filter/backdrop-filter carry a deterministic noise. What is missing is narrow, and it is what stops those primitives from composing into the shapes the reference reels build.

Owned files

• crates/rustmotion-components/src/shape.rs
• crates/rustmotion-core/src/schema/video.rs (Stroke)
• crates/rustmotion-core/src/css/style.rs (ClipPath node-reference variant only)

Do NOT write: engine/deps.rs (#328 owns the resolver this calls), expr/** (#327).

Deliverables

1. Stroke control on an arbitrary path. dash, dash_offset (animatable), line_cap and line_join exist only on arrow, connector and line today. Move them onto Stroke so shape.path gets them too.

2. path.data accepts an expression. This is what turns a gem into twenty-four triangular facets computed from eight outline points, an inner ring and an offset centre, each tinted by its angle to a light. Without it, procedural geometry stays impossible whatever else lands.

3. clip-path may reference another node's path. The reference reel's cracking stone is one gem drawn twice, each half clipped by a path following the crack, so the two halves separate along a coherent break. A rectangular overflow: hidden cannot express that and ClipPath::Path alone cannot either, because the crack is a node with its own animation.

Acceptance

• A for-each over 24 facets with computed data and computed fill renders a shaded gem
• A dashed arbitrary path animates its dash_offset into a draw-on
• Two shapes clipped by the same referenced crack path separate without a seam

[LeadcodeDev]

Two of the three acceptance criteria landed at 187cf7c (#340):

• computed data and computed fill in a for-each — the ${...} marker dialect in shape.rs
• stroke.dash_offset as a Computed<f32> resolved per frame (resolve_dash_offset, shape.rs:87), plus stroke-linecap / stroke-linejoin (schema/video.rs:1099)

The third — two shapes clipped by the same referenced crack path — did not. ClipPath::NodePath is declared and serialized (css/style.rs:1607) and its own doc comment says resolving it is the paint pipeline's job. That job was never done, and #339 measures the wider fault: no ClipPath variant is read by anything, NodePath included. Remainder tracked there rather than kept open here.