feat(animation): burst throws a ring of strokes off a node's box edge

[LeadcodeDev]

Closes #379 (second half — emitter shipped in #407).

The gap

A pill that pops needs a splash of ink to sell the pop. Today that is N hand-placed
line nodes at computed angles, each with its own draw_in timing.

{ "name": "burst", "delay": 0.4, "duration": 0.6, "count": 12, "length": 60,
  "gap": 16, "width": 5, "color": "#FFB020", "seed": 3, "jitter": 0.3 }

The stroke

Each stroke is a head and a tail that each cross the same track once: the head
runs out over the first half of the window (ease_out), the tail follows over the
second (ease_in). The stroke reaches full length at the halfway point and then
closes outward — it flies away, it does not retract into the box.

So the stroke has zero length at both ends of the window by construction, not
because a fade got small.

Why that geometric zero is not redundant with the window test

burst_progress does short-circuit outside [delay, delay + duration). But that
test cannot be exact for every duration: delay + duration - delay != duration in
floating point as soon as duration is not representable — 0.4 is the obvious
case — so a frame landing exactly on the end can fall one ULP inside the window.
It paints nothing anyway, because tail == head there.

a_frame_landing_one_ulp_inside_the_window_still_paints_nothing pins that, and it
is the reason the pixel-identity test alone is not enough: reverting the window
short-circuit to a .clamp(0.0, 1.0) fails
nothing_before_the_delay_and_nothing_at_or_after_the_end but leaves the pixel test
green, exactly as it should.

gap follows the shape

The track starts where the ray leaves the box — measured against the node's own
half-extents — plus gap. A wide pill therefore pushes its horizontal strokes
further out than its vertical ones, which is what makes the ring read as belonging
to the shape rather than sitting on a circle around it. Removing gap fails
burst_leaves_the_node_itself_untouched.

Determinism

Angle, per-stroke length and per-stroke phase all come from (seed, index) through
the same splitmix64-style hash shatter uses. Injecting SystemTime::now() into it
fails burst_two_renders_of_the_same_instant_are_byte_identical immediately.

Placement

Strokes are painted over the node, after its own render and inside its transform,
so they follow a pop_in or a rotation, and they take no layout space. They can
leave the viewport if the node sits near an edge — the geometry validator does not
see them, since a burst has no layout box; rules/burst.md says to leave
gap + length of margin.

Budget-checked like shatter and chromatic_aberration.

Reverts that bite

Revert	Test that goes red
drop gap from the track start	burst_leaves_the_node_itself_untouched
spacing fixed at TAU/8 instead of TAU/count	burst_strokes_sit_on_the_rays_count_asks_for_and_nowhere_else
SystemTime in the angle hash	burst_two_renders_of_the_same_instant_are_byte_identical
window short-circuit → .clamp(0.0, 1.0)	nothing_before_the_delay_and_nothing_at_or_after_the_end

Gate

cargo fmt --all --check clean · cargo clippy --workspace --all-targets --features rustmotion/studio -D warnings clean · cargo test --workspace 1767 passed · CLI probe validates and renders the ring above.