perf: Optimize vector viz search performance

- Replace sequential Qdrant scroll calls with batch retrieve
  (50 HTTP requests → 1 request, ~50x faster vector fetch)

- Add point_id to SearchResult to enable batch retrieval by Qdrant point ID

- Reuse query embedding from search algorithm in viz_routes
  (eliminates redundant embedding call, saves ~30ms)

- Make BM25 encode() async with thread pool to avoid blocking event loop
  (~4.4s was blocking, now properly async)

- Run PCA computation in thread pool to avoid blocking event loop
  (~1.2s was blocking, now properly async)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <noreply@anthropic.com>

nextcloud_mcp_server/embedding/bm25_provider.py

@@ -37,7 +37,9 @@ class BM25SparseEmbeddingProvider:
 
     def encode(self, text: str) -> dict[str, Any]:
         """
-        Generate BM25 sparse embedding for a single text.
+        Generate BM25 sparse embedding for a single text (synchronous).
+
+        Note: For async contexts, prefer encode_async() to avoid blocking the event loop.
 
         Args:
             text: Input text to encode
@@ -53,6 +55,23 @@ class BM25SparseEmbeddingProvider:
             "values": sparse_embedding.values.tolist(),
         }
 
+    async def encode_async(self, text: str) -> dict[str, Any]:
+        """
+        Generate BM25 sparse embedding for a single text (async).
+
+        Runs CPU-bound BM25 encoding in thread pool to avoid blocking the event loop.
+
+        Args:
+            text: Input text to encode
+
+        Returns:
+            Dictionary with 'indices' and 'values' keys for Qdrant sparse vector
+        """
+        import anyio
+
+        # Run CPU-bound BM25 encoding in thread pool
+        return await anyio.to_thread.run_sync(lambda: self.encode(text))  # type: ignore[attr-defined]
+
     async def encode_batch(self, texts: list[str]) -> list[dict[str, Any]]:
         """
         Generate BM25 sparse embeddings for multiple texts (batched).