[SYSTEMDS-3887] Add video modality to Scuro optimizer pipeline

src/main/python/systemds/scuro/dataloader/video_loader.py

@@ -37,6 +37,7 @@ class VideoStats:
     max_height: int
     max_channels: int
     num_instances: int
+    num_total_instances: int
 
     @property
     def output_shape(self):
@@ -132,8 +133,20 @@ def get_stats(self, source_path: str):
             max_height = max(max_height, height)
             max_num_channels = max(max_num_channels, num_channels)
             num_instances += 1
+        num_total_instances = num_instances
+        num_instances = (
+            min(num_instances, self.chunk_size)
+            if self.chunk_size is not None
+            else num_instances
+        )
         return VideoStats(
-            fps, max_length, max_width, max_height, max_num_channels, num_instances
+            fps,
+            max_length,
+            max_width,
+            max_height,
+            max_num_channels,
+            num_instances,
+            num_total_instances,
         )
 
     def estimate_peak_memory_bytes(self) -> dict:

src/main/python/systemds/scuro/drsearch/modality_shared_memory.py

@@ -401,3 +401,26 @@ def add_shared_memory_candidate(data: Any, resident_bytes: int = 0) -> bool:
             return data, shm.name, data_nbytes, resident_bytes
 
     return None, None, 0, resident_bytes
+
+
+_SHARED_MEMORY_WRAPPER_TYPES = (
+    SharedStringList,
+    SharedGroupedArrayList,
+    SharedArrayList,
+    SharedNDArray,
+)
+
+
+def collect_shm_names_from_payload(data: Any) -> List[str]:
+    if data is None:
+        return []
+    if isinstance(data, _SHARED_MEMORY_WRAPPER_TYPES):
+        return [data.shm_name]
+    if hasattr(data, "data"):
+        return collect_shm_names_from_payload(data.data)
+    if isinstance(data, (list, tuple)):
+        names: List[str] = []
+        for item in data:
+            names.extend(collect_shm_names_from_payload(item))
+        return names
+    return []