Keep referenced Subjects as stubs instead of deleting them

[JeroenDeDauw]

Fixes #544

savePage() DETACH DELETEd Subjects that were removed from a page without checking whether other Subjects still held relations to them, silently breaking those relations. deleteSubject() (used by deletePage) already guarded against this via subjectHasIncomingRelations(). savePage() now reuses deleteSubject(), so a removed-but-still-referenced Subject is preserved rather than deleted.

This also unifies the three kinds of "referenced but absent" node onto a single stub shape: a node with only the id and wiki_id properties and the Subject label (no name, no Schema label, no other properties). The paths:

• deleteSubject()'s keep path now strips the node to a stub, resolving the clear-properties / clear-labels TODOs (it previously kept the full stale node).
• savePage()'s new keep path, via deleteSubject().
• The placeholder MERGEd by Neo4jSubjectRelationUpdater when a relation targets a not-yet-existing Subject now gets the Subject label and wiki_id on create (it was previously a bare, unlabelled node).

When the real Subject is later saved, the normal save path upgrades the stub in place. All the MERGEs that reach these nodes match by id alone, so both the new labelled stubs and any pre-existing bare placeholders upgrade without creating a duplicate node.

Placeholder wiki_id is stamped with the current wiki's wiki_id, which is correct under Subject Sources v1 where relation targets are restricted to resolvable local sources. The graph is a rebuildable projection and NeoWiki is not in production, so any lingering old-shape placeholder nodes are acceptable; a graph rebuild normalises them and the id-based upgrade path handles them regardless.

docs/api/graph-model.md documents the stub shape.

Covered by integration tests (@group Database): savePage keeping a still-referenced removed Subject as a stub (regression), savePage still fully deleting an unreferenced removed Subject, deleteSubject producing a stub with cleared properties and Schema labels, the relation placeholder carrying the Subject label and wiki_id, and a real-Subject save upgrading a stub in place without duplication.

_{AI-authored — Claude Code, Opus 4.8 (max); batch execution of the Relations Phase 1 plan directed by @JeroenDeDauw, autonomous subagent run (no questions asked); diff not yet human-reviewed; TDD regression seen red then green, all new tests confirmed failing on reverted production, full PHPUnit suite (1798 tests) and make cs (phpcs + phpstan) green locally, CI green.}

Review notes (batch pr-review pass)

A fresh-session review pass added the missing stub coverage (63cdc49): a 2-incoming×2-outgoing stub test proving outgoing relations and statement-derived properties are stripped while incoming relations survive, teeth-verified by mutating the production clause and watching it fail. Known limitation, accepted for now: removing two mutually-referencing subjects in one save can leave an orphan stub (order-dependent; harmless — no reader surfaces it — and self-healing on RebuildGraphDatabases.php). A robust fix needs cascade sweeping, so it is left for a follow-up.

Accuracy correction to the attribution note above: two of the original tests (…UnreferencedSubjectDeletesIt, …UpgradesItsStubInPlace) are guard tests that also pass on master; the revert-verification claim holds for the four fix-proving tests.

_{Review-pass notes added by Claude Code, Fable 5 (max), summarizing a fresh-session Opus 4.8 review agent's verified findings; not yet human-reviewed.}