[onert/python] Add train flag to train_step

This commit adds `train` flag to train_step to toggle weight updates.
- Introduce `train: bool = True` parameter in `train_step`
- Pass `update_weights=train` to `session.train()`

ONE-DCO-1.0-Signed-off-by: ragmani <ragmani0216@gmail.com>

Author: ragmani

runtime/onert/api/python/package/experimental/train/session.py

@@ -246,14 +246,17 @@ def _check_batch_size(self,
                     f"{data_type} batch size mismatch at index {idx}: "
                     f"shape[0] = {arr.shape[0]} vs batch size = {batch_size}")
 
-    def train_step(self, inputs: List[np.ndarray],
-                   expecteds: List[np.ndarray]) -> Dict[str, Any]:
+    def train_step(self,
+                   inputs: List[np.ndarray],
+                   expecteds: List[np.ndarray],
+                   train: bool = True) -> Dict[str, Any]:
         """
-        Train the model for a single batch.
+        Train (or evaluate) the model for a single batch.
 
         Args:
             inputs (list of np.ndarray): List of input arrays for the batch.
             expecteds (list of np.ndarray): List of expected output arrays for the batch.
+            train (bool): If True, update weights; if False, run only forward & metrics.
 
         Returns:
             dict: Loss and metrics values, and train_time in ms.
@@ -282,7 +285,7 @@ def train_step(self, inputs: List[np.ndarray],
 
         # Run a single training step
         t_start: float = time.perf_counter()
-        self.session.train(update_weights=True)
+        self.session.train(update_weights=train)
         t_end: float = time.perf_counter()
 
         # Update loss