refactor: apply 5-pass audit optimizations across mcp-memory codebase

server/src/embedding.rs

@@ -33,9 +33,16 @@ pub fn cosine_similarity(a: &[f32], b: &[f32]) -> f32 {
     if a.is_empty() || b.is_empty() || a.len() != b.len() {
         return 0.0;
     }
-    let dot_product: f32 = a.iter().zip(b.iter()).map(|(x, y)| x * y).sum();
-    let norm_a: f32 = a.iter().map(|x| x * x).sum::<f32>().sqrt();
-    let norm_b: f32 = b.iter().map(|x| x * x).sum::<f32>().sqrt();
+    let mut dot_product = 0.0f32;
+    let mut norm_a_sq = 0.0f32;
+    let mut norm_b_sq = 0.0f32;
+    for (&x, &y) in a.iter().zip(b.iter()) {
+        dot_product += x * y;
+        norm_a_sq += x * x;
+        norm_b_sq += y * y;
+    }
+    let norm_a = norm_a_sq.sqrt();
+    let norm_b = norm_b_sq.sqrt();
     if norm_a == 0.0 || norm_b == 0.0 {
         0.0
     } else {
@@ -49,8 +56,13 @@ pub async fn generate_embeddings_async(texts: Vec<String>) -> Result<Vec<Vec<f32
     tokio::task::spawn_blocking(move || {
         let model_mutex = get_embedding_model()?;
         let mut model = model_mutex.lock().map_err(|e| e.to_string())?;
-        let embeddings = model.embed(texts, None).map_err(|e| e.to_string())?;
-        Ok(embeddings)
+        let mut all_embeddings = Vec::with_capacity(texts.len());
+        for chunk in texts.chunks(32) {
+            let chunk_vec = chunk.to_vec();
+            let chunk_embeddings = model.embed(chunk_vec, None).map_err(|e| e.to_string())?;
+            all_embeddings.extend(chunk_embeddings);
+        }
+        Ok(all_embeddings)
     })
     .await
     .map_err(|e| e.to_string())?