[LeadcodeDev]

Closes #431. Part of #438.

Six items, five of them a declared property that rendered nothing.

draw_start on a finished stroke. Ignored on an svg whose stroke was
finished or static: the progress >= 1.0 branch delegated straight to resvg and
never consulted it. svg now folds draw_start, draw_progress and the new
draw_offset into one window, and takes the resvg shortcut only when that window
is the whole path — so a finished stroke can be permanently trimmed from its tail
by the same segment extraction used mid-reveal, with the cap, join and width it
already had.

arrow, connector, shape. The first two never read props.draw_start at
all; shape resolved only its own static field and ignored keyframes. All three
go through trim_path_between now.

The 8×8 dot at draw_progress: 0. Same fix, same cause line.rs already
solved: a dash of [draw_len, total_len - draw_len + 0.01] with draw_len = 0
and a hardcoded round cap paints a solid dot of diameter width. arrow and
connector copy line's PathMeasure extraction and return without painting on
an empty window.

draw_offset. A new AnimatedProperties field shifting both ends of the
reveal window, each clamped to [0, 1] — a comet that marches and shrinks at the
far end rather than wrapping, since trim_path_between has no two-segment model.
Animatable by keyframes and wiggle.

d keyframes with target. These could not go through the generic animation
system: KeyframeValue is Number | Color, so a path string cannot round-trip
through it, and adding "d" to KNOWN_MOTION_PROPERTIES would have been exactly
the accepted-but-inert field this repository keeps closing. svg.path_morph
mirrors shape.path_morph instead, with target naming an element id because a
document usually holds more than one path. An unmatched target falls back to the
untouched document rather than blanking the frame.

Structure mismatch: refused, not resampled. Resampling arbitrary SVG
topologies is a large piece of work that produces silently ugly interpolations,
and the repository's taste is the named refusal — the draw: true rejection is
the direct precedent. The check reuses interpolate_path_data's own mismatch
detection rather than reimplementing path math.

Known gaps, stated rather than left to be found

• draw_offset has no effect on line. line.rs was read-only for this work
  — it is the module the others were imitating. Wiring it in is a follow-up.
• path_morph.target matching is a substring scan for id="…", in both the
  render-time patcher and the validator. Enough for a plain <path id="…">, and
  it would miss a namespaced attribute or unusual whitespace. A real fix parses
  with usvg/roxmltree at both ends.

Tests

Render tests with pixel assertions throughout — every item reverted individually
and watched go red:

draw_progress_zero_paints_nothing_not_even_a_dot   got 124 lit pixels
draw_start_erases_the_head_of_an_already_finished  left: 255 right: 0
draw_offset_marches_the_drawn_window_along_...     left: 255 right: 0
draw_offset_is_a_recognized_keyframe_property      "unknown animation property 'draw_offset'"

cargo test -p rustmotion --lib cli::commands::validate_schema 47 passed. Clippy
and the full suite run at integration; a sibling workstream is still mid-flight.