fix(cli): count font axes as motion in the sweep fingerprint

[miguel-heygen]

What

Fold the computed font-variation-settings into the layout audit's per-element sweep fingerprint.

Why

The fingerprint is rect plus opacity, with an 8x8 pixel hash for <canvas> and <video>. A composition whose only change across the timeline is a font axis touches none of those, so all six samples hash identically and check reports sweep_static on a timeline that is visibly animating.

Duplexed variable fonts make this unavoidable rather than unlikely. Recursive holds a single advance width across its entire weight axis by design, so even the element rect stays fixed while the letterforms change underneath it.

This is the third instance of the same failure class in this function. The canvas-repaint and opacity-reveal cases already have regression tests sitting right where these two go, which is itself the argument that the fingerprint is the correct place to fix it rather than the compositions.

How

font-variation-settings reads "normal" on every element that does not use it, so no existing composition's fingerprint changes.

Test plan

• Unit tests added/updated
• Manual testing performed
• Documentation updated (if applicable)

packages/cli full suite: 175 files passed, 1 skipped, 2489 tests passed, 2 skipped, 0 failed. oxfmt --check and oxlint clean on both files.

Two regression tests, both mutation-checked rather than trusted green:

Sabotage	Failing test
Drop the axis string from the fingerprint	changes the sweep fingerprint when only font-variation-settings moves
Add Math.random() to the axis read	keeps the sweep fingerprint identical when nothing moves, font axes included

Exactly one test failed in each case (1 failed, 82 passed). The second is the load-bearing one: without it a fingerprint that varied on its own would make sweep_static unfireable and every green layout verdict meaningless, and the suite would still be green.

Worth knowing

A fresh worktree needs bun run build before vitest, or 68 files fail to collect on Cannot find package '@hyperframes/parsers/ff-binaries'. That is missing workspace build output, not a real failure, and it looks alarming.

feat/primitive-weight-wave depends on this landing first. Demonstrated on the same composition: the CLI without this fix reports sweep_static and check fails; with it, layout reports 0 issues across 9 samples and check passes.