migrate article embeddings to support multi-model architecture and enhance data integrity

src/embeddings.js

@@ -111,6 +111,29 @@ function rebuildVec0IfModelChanged() {
     SELECT 1 FROM article_embedding_meta WHERE model != ? LIMIT 1
   `).get(EMBEDDING_MODEL);
 
+  // also sync any store entries that didn't make it into vec0
+  const missing = db.prepare(`
+    SELECT article_id, embedding FROM article_embedding_store
+    WHERE model = ?
+      AND NOT EXISTS (SELECT 1 FROM article_embeddings WHERE article_id = article_embedding_store.article_id)
+  `).all(EMBEDDING_MODEL);
+
+  if (missing.length > 0) {
+    const insertVec = db.prepare(`INSERT OR IGNORE INTO article_embeddings (article_id, embedding) VALUES (?, ?)`);
+    const insertMeta = db.prepare(`INSERT OR IGNORE INTO article_embedding_meta (article_id, model) VALUES (?, ?)`);
+
+    const sync = db.transaction(() => {
+      for (const row of missing) {
+        const vals = new Float32Array(row.embedding.buffer, row.embedding.byteOffset, row.embedding.byteLength / 4);
+        insertVec.run(BigInt(row.article_id), padEmbeddingForVec0(vals));
+        insertMeta.run(row.article_id, EMBEDDING_MODEL);
+      }
+    });
+
+    sync();
+    console.log(`synced ${missing.length} store embeddings into vec0`);
+  }
+
   if (!stale) return;
 
   const storeCount = db.prepare(`