refactor: deduplicate articles at insertion time instead of query time

Change deduplication strategy from query-time (ROW_NUMBER window function)
to insertion-time (global guid check before insert).

Benefits:
- Simpler queries without CTE/window functions
- Consistent read state (no duplicate articles to manage)
- Better query performance (no per-query deduplication overhead)

Changes:
- Add CheckArticleExistsByGUID query for global guid lookup
- Add migration to remove existing duplicate articles
- Modify fetchOneFeed and AddFeed to skip duplicates on insert
- Revert GetUnreadArticles/GetReadArticles to simple queries

1 files changed, 9 insertions, 1 deletions

diff --git a/backend/db/migrations/005_add_guid_index.sql b/backend/db/migrations/005_add_guid_index.sql
index a653d79..e3625ee 100644
--- a/backend/db/migrations/005_add_guid_index.sql
+++ b/backend/db/migrations/005_add_guid_index.sql
@@ -1,2 +1,10 @@
--- Add index on guid for deduplication queries
+-- Add index on guid for deduplication
 CREATE INDEX IF NOT EXISTS idx_articles_guid ON articles(guid);
+
+-- Remove duplicate articles by guid, keeping only the one with the smallest id
+DELETE FROM articles
+WHERE id NOT IN (
+    SELECT MIN(id)
+    FROM articles
+    GROUP BY guid
+);