ActionList: multiselect checkmark animation ignores prefers-reduced-motion #8648

Component

ActionList (selectionVariant="multiple")

Description

The checkbox that ActionList renders for multi-select items wipes its checkmark in and out with a clip-path animation, and that animation runs even when the visitor has asked for reduced motion.

.MultiSelectCheckbox::before in packages/react/src/ActionList/ActionList.module.css applies:

transition: visibility 0s linear 230ms;
clip-path: inset(var(--base-size-16) 0 0 0);
animation: checkmarkOut 80ms cubic-bezier(0.65, 0, 0.35, 1);

and the checked state adds animation: checkmarkIn 80ms ... forwards 80ms. Nothing here is wrapped in a media query, so prefers-reduced-motion: reduce gets the same wipe as everyone else.

Most of the library opts in to motion instead of opting out — Skeleton/SkeletonBox.module.css, SegmentedControl, ProgressBar, TooltipV2 and Breadcrumbs all put their animation declarations inside @media (prefers-reduced-motion: no-preference). Checkbox has the identical wipe and is reported separately in #8646.

Steps to reproduce

1. Open the FilteredActionList > features > MultiSelectWithSelectAll story (it renders ActionList with selectionVariant="multiple", which is what produces .MultiSelectCheckbox).
2. Turn on prefers-reduced-motion: reduce — in Chrome DevTools: Rendering → Emulate CSS media feature prefers-reduced-motion.
3. Click an item and watch the mark.

Expected: the mark appears and disappears immediately.
Actual: the 80 ms wipe plus the 230 ms delayed visibility still run.

Proposed fix

Keep the end state always visible and move only the animation into the media query, so the reduced-motion path renders the same pixels the animated path lands on:

• base ::before: clip-path: inset(0 0 0 0), with the visibility delay and checkmarkOut inside @media (prefers-reduced-motion: no-preference)
• checked ::before: un-clipped by default, with the clipped start value and checkmarkIn inside the same media query

That is the shape proposed in #8647 for Checkbox, and it leaves the animated result untouched, so the existing visual regression baselines should stay valid.

I'd like to open a PR for this — happy to follow whatever the Checkbox one settles on so both land the same way.