feat(clip-path): a morph can pass through intermediate shapes

[LeadcodeDev]

Closes #384. Refs #388.

The decision, and why it is neither option I offered

#398 landed clip-path animation in a different spelling from the one this issue describes, and I left it open with the choice written into the issue: finish the literal property: "clip-path" form with shape-valued keyframes, or re-scope.

On reflection, neither. The spelling was never the gap — the limit to two shapes was.

A second vocabulary for one capability is worse than one good one, and a shape-valued keyframe would have to re-earn the easing, spring, loop and per-keyframe easing that the scalar clip_path_progress track already provides for free, precisely because it is an ordinary motion property.

So the chain gets longer instead:

"clip-path": {
  "kind": "morph",
  "from": { "kind": "circle", "radius": 20 },
  "via":  [ { "kind": "circle", "radius": 140 } ],
  "to":   { "kind": "circle", "radius": 60 }
}

How it behaves

progress spans the whole chain, not each leg: with one via, 0.5 lands exactly on it and 0.25 is halfway along the first leg. Legs are equal in progress whatever the geometric distance between the shapes — an uneven pace is the easing's job, and putting it in via would have made two knobs fight over the same thing.

Each consecutive pair follows the same rule as a two-shape morph: same kind, same vertex count for a polygon. An incompatible pair mid-chain warns on stderr and leaves the node unclipped, exactly as a lone incompatible pair does — no new failure mode.

Verification

Four tests. Three fail without the chain, confirmed by short-circuiting morph_leg:

a_morph_chain_passes_through   FAILED
a_morph_chain_keeps_moving     FAILED
a_via_shape_splits             FAILED
an_empty_via_leaves            ok        ← the compatibility pin, correctly

an_empty_via_leaves_the_two_shape_morph_exactly_as_it_was asserts on morph_leg directly rather than through pixels: with no via, the endpoints and the progress pass through untouched. That is the clearest possible statement that nothing written before this renders differently.

One test replaced an earlier version of itself that was simply wrong. It compared a chain at progress 1.0 against a direct morph at 0.5 — which land on different shapes by construction. It asserted something untrue, and failed for the right reason. Worth saying, because a test that fails for a bad reason is easy to "fix" by loosening it.

cargo fmt --all --check, cargo clippy --workspace --all-targets -- -D warnings, cargo test --workspace (1722) all clean.

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