perf(#12): push suggest() filtering + LIMIT to SQL

Replace the in-memory approach (list_files → filter in Rust) with
SQL-level ILIKE + relevance ORDER BY + LIMIT.

- Add suggest_files_by_name() to FileReadPort (default impl for stubs)
- Add suggest_folders_by_name() to FolderRepository (default impl)
- Implement efficient SQL in FileBlobReadRepository & FolderDbRepository
  with LOWER(name) LIKE pattern, 3-tier relevance sorting, and LIMIT
- Rewrite SearchService::suggest() to call new methods in parallel

Before: 50K files → ~10 MB transferred, 52K string comparisons
After:  50K files → ~20 rows transferred, index-assisted scan

src/domain/repositories/folder_repository.rs

@@ -120,4 +120,27 @@ pub trait FolderRepository: Send + Sync + 'static {
         let _ = (folder_id, name_contains, user_id);
         Ok(Vec::new())
     }
+
+    /// Return up to `limit` folders whose name contains `query` (case-insensitive).
+    ///
+    /// Results are ordered by relevance (exact > starts-with > contains) for
+    /// autocomplete suggestions.
+    ///
+    /// The default implementation falls back to `list_folders` + in-memory
+    /// filter so that stubs and mocks compile without changes.
+    async fn suggest_folders_by_name(
+        &self,
+        parent_id: Option<&str>,
+        query: &str,
+        limit: usize,
+    ) -> Result<Vec<Folder>, DomainError> {
+        let all = self.list_folders(parent_id).await?;
+        let q = query.to_lowercase();
+        let mut matched: Vec<Folder> = all
+            .into_iter()
+            .filter(|f| f.name().to_lowercase().contains(&q))
+            .collect();
+        matched.truncate(limit);
+        Ok(matched)
+    }
 }