Release v1.52.0-next.2

@backstage/backend-defaults@0.17.3-next.2

89a95ca: Fixed the task worker retry loop to respect the abort signal. Previously, when a task worker encountered an unexpected error, the retry loop would continue indefinitely even after the worker was signaled to stop. The retry loop now checks the abort signal before retrying and passes it to the retry delay, allowing the worker to shut down gracefully.

c161e1c: Lazy-load react-syntax-highlighter and @dagrejs/dagre so they are no longer pulled in eagerly through the barrel export. This reduces the upfront module cost of importing from @backstage/core-components by roughly 10 MB. The public API is unchanged.

9698738: Dropped the legacy search_entity_id_idx index which is now redundant with the covering unique index on (entity_id, key, value). The old index caused the query planner to choose an inefficient scan pattern for catalog list queries with multiple sort fields, leading to severely degraded performance on large catalogs.
ccfa4f1: Optimized entitiesBatch on PostgreSQL to use = ANY(array) instead of WHERE IN ($1, $2, ...). This produces a single stable query plan regardless of batch size, instead of up to 200 different plans that pollute the query plan cache. On PostgreSQL, batching is no longer needed so all entity refs are fetched in a single query.
24775dc: Added a migration that tunes PostgreSQL automatic vacuum thresholds on the search, final_entities, relations, and refresh_state_references tables, and fixes column statistics for entity_id in the search table. This prevents the query planner from falling back to sequential scans when table maintenance falls behind, keeping catalog queries fast on large installations.

e6493b9: Fixed a bug where setting user.select to an empty array would cause all users to be dropped from the catalog instead of using the default set of fields.