feat(supabase): sync knowledge documents and prune tombstones daily

knowledge_documents deletions now leave a sync tombstone and have a cursor
index and Realtime, so the desktop mirror can follow them. pg_cron runs
prune_sync_tombstones_v1 every day at 03:17 UTC with the 180-day window.

server/supabase/migrations/20260927000035_sync_knowledge_and_retention.sql

@@ -0,0 +1,55 @@
+-- Knowledge base joins the cross-device mirror; tombstones get a retention job.
+--
+-- 1. knowledge_documents deletions leave a tombstone like the other synced
+--    tables, and a (user_id, updated_at, id) cursor index backs keyset pulls.
+--    Chunks are re-read with their document, so they need neither.
+-- 2. pg_cron prunes tombstones daily with the 180-day default of
+--    prune_sync_tombstones_v1 (a device offline longer than that resyncs fully).
+
+BEGIN;
+
+-- 1) knowledge_documents ---------------------------------------------------------
+ALTER TABLE public.sync_tombstones DROP CONSTRAINT IF EXISTS sync_tombstones_table_name_check;
+ALTER TABLE public.sync_tombstones
+  ADD CONSTRAINT sync_tombstones_table_name_check CHECK (table_name IN (
+    'history',
+    'dictionary',
+    'meetings',
+    'meeting_memos',
+    'meeting_documents',
+    'memo_tags',
+    'custom_instructions',
+    'user_templates',
+    'knowledge_documents'
+  ));
+
+DROP TRIGGER IF EXISTS record_sync_tombstone_v1 ON public.knowledge_documents;
+CREATE TRIGGER record_sync_tombstone_v1
+  AFTER DELETE ON public.knowledge_documents
+  FOR EACH ROW EXECUTE FUNCTION public.record_sync_tombstone_v1();
+
+CREATE INDEX IF NOT EXISTS idx_knowledge_documents_sync_cursor
+  ON public.knowledge_documents(user_id, updated_at, id);
+
+DO $$
+BEGIN
+  IF NOT EXISTS (
+    SELECT 1 FROM pg_publication_tables
+    WHERE pubname = 'supabase_realtime' AND schemaname = 'public' AND tablename = 'knowledge_documents'
+  ) THEN
+    ALTER PUBLICATION supabase_realtime ADD TABLE public.knowledge_documents;
+  END IF;
+END $$;
+
+COMMIT;
+
+-- 2) Tombstone retention ---------------------------------------------------------
+-- Outside the transaction: extension creation and cron scheduling are
+-- idempotent, and cron.schedule with an existing job name replaces it.
+CREATE EXTENSION IF NOT EXISTS pg_cron WITH SCHEMA pg_catalog;
+
+SELECT cron.schedule(
+  'prune-sync-tombstones',
+  '17 3 * * *',
+  $$SELECT public.prune_sync_tombstones_v1()$$
+);