fix(Modal,Slideover): render the actions slot when nothing else opens the header

[IgorShevchik]

Linked issue

Refs #87 (first half — attaching to the document. The 27 MB corpus itself is not shrunk here; that is the second half and gets its own PR.)

Follow-ups filed from this work: #513 (293 hand-written mountSuspended calls are still detached) and #514 (the console gate's register is all-or-nothing per file).

Type of change

• Documentation (updates to the documentation or readme)
• Bug fix (a non-breaking change that fixes an issue)
• Enhancement (improving an existing functionality)
• New feature (a non-breaking change that adds functionality)
• Chore (updates to the build process or auxiliary tools and libraries)
• Revert (undoing a merged change — retitle this PR revert(Scope): ...)
• Breaking change (fix or feature that would cause existing functionality to change)

Description

componentRender now mounts with attachTo: document.body and unmounts as soon as it has the markup. Both halves were doing work the suite could not see.

Attaching. reka-ui's dialogs check their own accessibility with document.getElementById(titleId) in onMounted. Against a detached tree that lookup fails, so every dialog spec rendering with portal: false warned that DialogContent requires a DialogTitle — with the title sitting in the markup it had just produced. Measured with the register in console-gate.ts emptied on both sides: 623 console messages from 62 tests before, 366 from 56 after.

It also moved 402 snapshot entries across 25 files toward what a browser renders: Header and Table gain the aria-hidden a browser puts on everything behind an open dialog, and CheckboxGroup/RadioGroup gain the aria-label reka-ui reads off the associated <label> rather than the raw value.

Unmounting is the half that found defects, because teardown had never run.

	What was wrong
Modal, Slideover	The actions slot was dropped whenever nothing else asked for a header. Slideover hit it on its own defaults — side: 'bottom' puts the close button outside the header, so <B24Slideover :actions> rendered nothing at all. Modal had the same hole twice, in the header and in contentWrapper, reachable with :close="false". Drawer had it right and is the reference.
Countdown	The requestAnimationFrame handle lived in a ref, which deep-wraps an object value in reactive(). A browser returns a number there; happy-dom returns the Node Immediate it scheduled, so clearImmediate was handed a Proxy and unlinked a node that was in no queue — four uncaught TypeError: Cannot read properties of undefined (reading '_idleNext') the moment pause() first ran. Now a plain let, dropped as it is spent.
test/utils/mount.ts	The vue project's router is a module singleton, and vue-router's install replaces app.unmount with one that resets currentRoute to START_LOCATION. Every mount after the first rendered against an un-navigated router, and Link to="/" silently stopped being active. The shim awaits router.isReady().

Three tests were asserting nothing and now assert something. Countdown's pauses counting never started the countdown, so it only watched cancelAnimationFrame be handed the initial 0. The two editor specs mocked registerPlugin without unregisterPlugin, which mattered only once teardown reached tiptap's beforeUnmount.

How the Modal/Slideover bug was found. The #454 collision guard: renders with actions slot was byte-identical to renders with open. The same guard also showed that four of its baseline groups were held apart by pointer-events: auto — which reka-ui renders from a module-global layer stack, so it recorded how many earlier cases were still mounted rather than anything under test. The baseline is regenerated and is smaller: 291 groups, 875 entries. Slideover never entered it, because the case was fixed instead of recorded.

One register entry goes: nuxt:components/Header.spec.ts. The rest stay because the register keys on files, and every dialog spec also mounts by hand outside componentRender — see #513.

Review

A second commit acts on review. continueProcess cancelled the rAF handle without dropping it, so the delay <= 0 branch carried an already-cancelled handle into stop() and cancelled it twice — the same shape the comment two lines above blames for the four TypeErrors. And the JSDoc on componentRender had grown to 33 lines against this repo's own twenty-line ceiling; the measurements and the note about mocks needing their teardown half moved to .github/contributing/testing.md.

One flagged item was left alone deliberately. DropdownMenu's and Select's with items cases are not newly vacuous — with items is the reference render of its group (29 KB, eleven real menu items), and what disappeared was one leaked pointer-events: auto byte that recorded how many earlier cases were still mounted. dropdown-menu's theme puts colour on the trigger, which these fixtures do not render, so this is the collision guard's own documented example of an unavoidable one.

Gate

lint, typecheck, 318 test files / 7474 tests, test:module, and coverage 71.98 / 69.99 / 71.18 / 71.55 against thresholds of 70 / 68 / 70 / 70.

Checklist

• I have linked an issue or discussion.
• I have updated the documentation accordingly.