⚡ Bolt: Use .all() for faster boolean reduction in NumPy (#282)

Co-authored-by: google-labs-jules[bot] <161369871+google-labs-jules[bot]@users.noreply.github.com>

Author: n24q02m

qwen3_embed/common/utils.py

@@ -30,7 +30,8 @@ def last_token_pool(input_array: NumpyArray, attention_mask: NDArray[np.int64])
     Returns:
         Pooled embeddings, shape (batch_size, hidden_dim).
     """
-    left_padding = bool(attention_mask[:, -1].sum() == attention_mask.shape[0])
+    # ⚡ Bolt: Fast boolean reduction using .all() (~15% faster than .sum() == shape[0])
+    left_padding = bool(attention_mask[:, -1].all())
     if left_padding:
         return input_array[:, -1]