Fix episode filtering bug when requesting a subset of the episodes in a dataset (#2456)

* filter episodes in load_nested_dataset

* nit

* remove test filtering

* move import to module level

* added missing episode indices to the EpisodeAwareSampler in lerobot_train.py;

src/lerobot/datasets/lerobot_dataset.py

@@ -830,7 +830,7 @@ class LeRobotDataset(torch.utils.data.Dataset):
     def load_hf_dataset(self) -> datasets.Dataset:
         """hf_dataset contains all the observations, states, actions, rewards, etc."""
         features = get_hf_features_from_features(self.features)
-        hf_dataset = load_nested_dataset(self.root / "data", features=features)
+        hf_dataset = load_nested_dataset(self.root / "data", features=features, episodes=self.episodes)
         hf_dataset.set_transform(hf_transform_to_torch)
         return hf_dataset
 
@@ -847,10 +847,8 @@ class LeRobotDataset(torch.utils.data.Dataset):
 
         # Determine requested episodes
         if self.episodes is None:
-            # Requesting all episodes - check if we have all episodes from metadata
             requested_episodes = set(range(self.meta.total_episodes))
         else:
-            # Requesting specific episodes
             requested_episodes = set(self.episodes)
 
         # Check if all requested episodes are available in cached data