feat(animation): letter_spacing and draw_start become keyframe properties

[LeadcodeDev]

Closes #380. Closes #377. Refs #388.

Both halves were blocked on the same thing — a new field on AnimatedProperties, in a file another workstream owned while #363/#380 and #377 were in flight.

One of those agents explicitly refused to work around it, and was right to: adding letter_spacing to KNOWN_MOTION_PROPERTIES alone would have made it validate and stay completely inert, which is the exact defect #363 was fixing. It stopped and reported instead. This PR is the other side of that boundary, now that the file is free.

letter_spacing

{ "property": "letter_spacing",
  "keyframes": [{ "time": 0, "value": 0 }, { "time": 0.8, "value": 18 }],
  "easing": "ease_out" }

The intrinsic measurement keeps the static style.letter-spacing, so animating it does not re-run layout — the policy every layout property follows here. The consequence is real, and the rule file leads with it rather than hiding it: a text whose spacing opens overflows its box. Size the box for the final value, or accept the overflow knowingly.

Two mechanisms now share a name, which is worth knowing before it surprises someone: a timeline changing letter-spacing snaps and the validator warns about it; a keyframes track on letter_spacing interpolates. Different subsystems, deliberately different behaviour.

draw_start

draw_progress advances the head of a stroke; draw_start advances its tail. Together they are a window travelling along the path — a line that runs rather than one that grows. On its own it erases the beginning of a finished stroke.

line and svg both trim through PathMeasure::get_segment — the primitive #403 added — rather than a dash effect. That is what keeps a zero-length dash with a round cap from painting the stray dot #376 had to remove; reusing the dash trick here would have reintroduced it on a different component.

shape keeps its own component-level draw_start from #403, because it carries path_morph too and the two belong together. The rule file says which is which.

Verification

Five tests. Three fail without the wiring:

a keyframed letter_spacing must move the glyphs apart: tight=48px, loose=48px
assertion `left == right` failed: draw_start 0.5 must erase the first half

Two assert equality and pass both ways on purpose — an absent animated value must leave the declared one alone, which is why the field defaults to NaN rather than 0.

One probe deliberately stops six pixels short of the cut, with the reason written into the assertion: a round cap extends half the 8px stroke back past the trim. That is the cap working, not the trim failing — and it was worth writing down rather than quietly widening the tolerance.

cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace (1718) all clean.

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