feat(search): add document_chunks page/source columns + unique idx

migrations 279-281: page_start/end + source_type/chunker_version/source_hash/chunk_content_hash, legacy backfill (30,952 rows), unique (doc_id,source_type,chunker_version,chunk_index). PR-DocSrv-LargeDoc-Split-Markdown-1 commit 1.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

migrations/279_document_chunks_split_columns.sql

@@ -0,0 +1,10 @@
+-- PR-DocSrv-LargeDoc-Split-Markdown-1 (commit 1: schema)
+-- 통합 schema: LargeDoc(page_start/end) + Phase 3A(source_type/chunker_version/source_hash/chunk_content_hash) 선반영.
+-- Phase 3A 세션은 이 컬럼 위에서 동작 (schema 미터치). 분담 plan: brisk-paging-quokka.md
+ALTER TABLE document_chunks
+  ADD COLUMN IF NOT EXISTS page_start integer,
+  ADD COLUMN IF NOT EXISTS page_end integer,
+  ADD COLUMN IF NOT EXISTS source_type text,
+  ADD COLUMN IF NOT EXISTS chunker_version text,
+  ADD COLUMN IF NOT EXISTS source_hash text,
+  ADD COLUMN IF NOT EXISTS chunk_content_hash text;

migrations/280_document_chunks_legacy_backfill.sql

@@ -0,0 +1,8 @@
+-- 기존 row 안전 backfill = legacy/legacy (과분류 X; Phase 3A 가 정확 source_type 으로 재생성/정리).
+-- page_start/end 는 기존 단일 page 컬럼으로 (page NULL 이면 NULL 유지).
+UPDATE document_chunks SET
+  source_type     = COALESCE(source_type, 'legacy'),
+  chunker_version = COALESCE(chunker_version, 'legacy'),
+  page_start      = COALESCE(page_start, page),
+  page_end        = COALESCE(page_end, page)
+WHERE source_type IS NULL OR chunker_version IS NULL OR page_start IS NULL OR page_end IS NULL;

migrations/281_document_chunks_source_unique_idx.sql

@@ -0,0 +1,5 @@
+-- UNIQUE (doc_id, source_type, chunker_version, chunk_index).
+-- dup-check (doc_id, chunk_index)=0 확인 (2026-05-24) → legacy backfill 후 키 유효.
+-- 비-CONCURRENTLY: runner 트랜잭션 안전 + ~30K rows 규모 현실적.
+CREATE UNIQUE INDEX IF NOT EXISTS uq_document_chunks_source_version_index
+ON document_chunks (doc_id, source_type, chunker_version, chunk_index);