[TASK] Migrate HyperLinkNode/ReferenceNode constructor calls to array children form

[CybotTM]

Summary

Upstream phpDocumentor/guides#1161 deprecated the string-form first argument of HyperLinkNode::__construct() and the string-form second argument of ReferenceNode::__construct() in favor of an array of InlineNodeInterface instances. The BC bridge still works but emits a Doctrine\Deprecations trigger on every call. Pre-existing tech debt — not introduced by any recent bump.

This PR removes every deprecated call site (constructor + BCInlineNodeBehavior::getValue()) in the theme, then carries the upstream data model (children as list<InlineNodeInterface>) through the local code so private helpers and node ctors stop passing flat strings around. Independent of #1243 — branched from origin/main.

Commits

1. 665aa4b6 Mechanical migration to array-form constructors. Each call site passes either [] or [new PlainTextInlineNode($value)] — mirroring the BC bridge's internal wrapping. Behavior-preserving.

2. 233d9004 Adopt upstream style + getChildren() passthrough.

   • TextRoles: switch to upstream's compact truthy form $x ? [new PlainTextInlineNode($x)] : [] (as in upstream ReferenceTextRole, ApiClassTextRole, ReferenceRule, Markdown LinkParser). Where the source has a guaranteed-non-empty fallback, the ternary is dropped entirely.
   • NodeTransformers: stop calling deprecated BCInlineNodeBehavior::getValue() to flatten an inline tree just to rewrap it. Pass $node->getChildren() through directly. Bit-identical for plain-text content (entire TYPO3 corpus); preserves inline structure (emphasis, code, nested refs) for any future content. Removes the last remaining BCInlineNodeBehavior::getValue() call sites in the codebase.

3. b4f2ba88 Push the children-array form through private helpers and FileInlineNode's parent ctor.

   • Three TextRoles (ViewhelperTextRole, ViewhelperArgumentTextRole, FileTextRole) had private createNodeWithInterlink helpers that still took string|null $referenceName and did the wrap-as-PlainTextInlineNode dance internally. Wrap once at the upstream-imposed createNode boundary instead, so the private helpers operate on the upstream data model directly (array $children typed list<InlineNodeInterface> for the two Viewhelper roles; non-nullable string $fileLabel for FileTextRole, since FileInlineNode's public surface is intentionally out of scope).
   • FileInlineNode::__construct silenced a hidden [TASK]: Bump friendsofphp/php-cs-fixer from 3.93.0 to 3.93.1 #1161 deprecation trigger: it called the 4-arg BC form parent::__construct(type, ref, string, array) — the BC bridge sees a string as 3rd positional and fires the deprecation regardless of the array as 4th positional. Switch to the 3-arg form (type, ref, array).

Test plan

All run locally with PHP_BIN=php on host PHP 8.5 after each commit:

• make phpstan → [OK] No errors
• make test-unit → OK (83 tests, 183 assertions)
• make test-integration → Tests: 113, Assertions: 1131, Deprecations: 3, Skipped: 2 (the 3 deprecations are pre-existing PHP 8.5 curl_close warnings in vendor code, identical to origin/main)
• make test-rendertest → completed, no fixture diffs (53 + 35 + 4 + 4 SINGLEPAGE files)
• make test-docs → completed (14 SINGLEPAGE files)

No expected-fixture HTML changed across any of the three commits, confirming the migration is behavior-preserving.

[CybotTM]

Closing this voluntarily — on reflection, this PR is preparatory cleanup with no user-facing benefit at this point in time:

• Doctrine\Deprecations is not configured to trigger errors in this project (no Deprecation::enableWithTriggerError() call, no failOnDeprecation in phpunit.xml.dist). The [TASK]: Bump friendsofphp/php-cs-fixer from 3.93.0 to 3.93.1 #1161 deprecation triggers fire silently into a counter — there is no observable symptom in CI, logs, or runtime today.
• Upstream's migration is visibly mid-stream: the AbstractReferenceTextRole vs. CustomLinkTextRole family split forces the wrap to happen at different architectural levels in our two TextRole families, and ResolvedInventoryLink is still @internal. A more cohesive migration path may emerge from upstream later (e.g., Rector recipe, unified API, or a firm BC-bridge removal timeline).
• Maintainer review bandwidth is real cost; submitting a no-benefit PR for review trades that without compensation.

The branch is preserved on my fork (CybotTM/render-guides:feature/modernize-inline-node-ctors) and can be re-submitted as a fresh PR when upstream's direction is clearer or a BC-removal date is announced.

The companion PR #1243 — which actually fixes the rendering bug from #1236 — stays open and is unaffected by this closure. The two were always independent (different branches, both based on origin/main).

Sorry for the noise.