feat(pages): extend the application page model for grid pages

[rubenvdlinde]

First step of the page rework agreed in nextcloud#3075 (reply: pages move onto the existing application page model instead of views without a table). Backend only, no UI.

What changes

• Migration Version2400Date20261015000001: tables_contexts_page gets title and technical_name (both nullable, 200). tables_contexts_page_content gets type (default data), configuration (text), grid_x, grid_y, grid_width, grid_height (nullable), and node_rel_id becomes nullable. Idempotent. App version 2.4.0-dev.1 so installs run it.
• Page::TYPE_GRID, technical name check ^[a-z][a-z0-9_]*$; Page and PageContent serialize the new fields additively. The start page keeps startpage.
• ContextMapper: pages and content come from one query without the pc.node_rel_id = r.id OR pc.node_rel_id IS NULL filter, so content without a node no longer matches every node. Node relations of the found contexts come from a second query, one per call (chunked). Same response shape for existing data.
• PageContentHelper reads content for every API output (Context::jsonSerialize): no type is a data widget, no configuration gets the type defaults (GridWidgetTypes::defaults), rows without a position go full width (12), 5 high, below the lowest positioned row, stacked by order. orderFromPositions gives the order for any write of positions.
• The context import (importContext) now stores content without a node (it failed against the NOT NULL column), carries title, technical name (dropped when invalid or a duplicate), type, configuration and positions, and fills order from positions.

Deviation from the reply, on purpose

The reply proposed moving pc.node_rel_id = r.id into the join. With the nodes joined through content, a node that is on no page drops out of nodes, and findAllContainingNode (used by PermissionsService) then stops granting its permissions. Once grid pages can remove a data widget that happens. Loading the node relations in their own query keeps every node listed without multiplying rows. findAllContainingNode filters contexts by a subquery and keeps only the matching node, as before.

GridWidgetTypes comes over with the defaults only; schemas, sanitising and the endpoint follow in the API PR.

Technical names required: every page has a technical name, unique per application (tables_ctx_page_name_uq); Version2400Date20261015000002 backfills existing pages (start page dashboard, others from the title, else page_<id>, deduped with _2, _3), Version2400Date20261015000003 makes it NOT NULL, and TechnicalNameHelper is the one generator.

Tests

Run in a Nextcloud 34 container (PostgreSQL 16), one file per call:

• tests/unit/Db/ContextMapperTest.php (new, real DB): OK (7 tests, 36 assertions). Nodes + content, content without a node, a node on no page (also via findAllContainingNode), empty context, pages without content, findAll.
• tests/unit/Helper/PageContentHelperTest.php (new): OK (6 tests, 30 assertions)
• tests/unit/Service/ContextServiceTest.php: OK (4 tests, 31 assertions)
• PermissionsServiceTest, TablesMigratorTest, ShareReviewSourceTest, EntityVirtualPropertiesTest: all OK
• composer cs:check clean, psalm No errors found!, composer openapi no drift.
• Manual: created an application via api/2 and read it back: start page content shows type: data, defaults and position 0/0/12/5.

Inherited: none on touched lines.

Prepared in the Conduction fork for porting upstream by @remko48. Prepared with Claude Code.

[rubenvdlinde]

Superseded by #46 (rebuilt on the nextcloud#3100 model, see #13). Kept open for reference until Remko has ported.