fix(signals): a reawakened lazy memo rejoins its owner's chain

[brenelz]

Fixes #3554

What was broken

An auto-dispose memo (createMemo(fn, { lazy: true })) that went dormant was
spliced out of its owner's child chain and never relinked when a later read
reawakened it. Its next dormancy ran the splice again from outside the chain
and wrote a stale _nextSibling into parent._firstChild, orphaning every
sibling created after it: never disposed by the owner, still subscribed to
their sources, still running after the root's dispose(). Dev builds threw
the owner-chain-head invariant from #3552 mid-flush instead.

Root cause

unobserved() calls disposeChildren(node, true), which unlinks the node,
nulls _prevSibling and leaves _nextSibling intact so outer walks still
advance. The reawaken in prepareComputed only called recompute(comp, true),
whose flag wipe cleared REACTIVE_DISPOSED. The node was then live by flags
but outside the chain, so the next splice took the head branch and clobbered
the head with that stale pointer.

What changed

prepareComputed links the node back at the chain head before recomputing, so
"live" again implies "in the chain" and the owner's dispose() and reruns
reach a reawakened memo. The relink is skipped for a zombie, whose splice was
skipped and which still sits on the pending chain, and for a node whose parent
is already disposed, whose chain can never be drained again.

disposeChildren now detaches node._firstChild before its loop and points
each child's _prevSibling at itself, so a node that a cleanup links during a
held-children drain stays on the chain instead of being written back into the
head at its next dormancy.

The creation-time link in createOwner and setupComputedNode and the new
relink share one linkChild helper. The owner-chain-head invariant is
unchanged and pins the regression. The core minified floor moves 25,671 to
25,724, under the 25,750 cap, and is recorded in the treeshake ledger.

The alternative in the issue, skipping the splice for dormant nodes, was
rejected: it leaves disposed members in the live chain for every walker to
tolerate, retains every dormant lazy memo until its owner disposes, and
contradicts the chain invariant the #3552 test encodes.

Verification

cd packages/signals && npx vitest run tests/dormant-memo-chain-3554.test.ts
cd packages/signals && SIGNALS_TIER=observe npx vitest run tests/dormant-memo-chain-3554.test.ts
cd packages/signals && npx vitest run

packages/signals/tests/dormant-memo-chain-3554.test.ts adds four cases. Three
fail on pristine next and all four pass here, under the dev tier and under
SIGNALS_TIER=observe. Full suites: signals 3,468 passed, solid 654 passed,
web 834 passed.

Left open

Dormant off-chain nodes keep CONFIG_AUTO_DISPOSE and still recompute after
their owner dies, the pre-existing #3024 gap. This change neither widens nor
closes it.

A relinked node moves to the chain head, so sibling disposal order for a
reawakened memo is no longer creation order.

[changeset-bot]

🦋 Changeset detected

Latest commit: e4a7cb9

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 11 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
@solidjs/web	Patch
@solidjs/babel-plugin	Patch
@solidjs/compiler	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
solid-js	Patch
@solidjs/universal	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[codspeed]

Merging this PR will not alter performance

✅ 172 untouched benchmarks
⏩ 3 skipped benchmarks¹

---

_{Comparing brenelz:claude/issue-3554 (e4a7cb9) with next (be46a04)}

Footnotes

1. 3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩