chore(datasets): trim sampler comment and drop duplicate tests

Remove the verbose dataloader-guard comment and the two EpisodeAwareSampler tests
that duplicated existing validation/warning coverage (no coverage loss).

Co-authored-by: Cursor <cursoragent@cursor.com>

src/lerobot/scripts/lerobot_train.py

@@ -389,11 +389,7 @@ def train(cfg: TrainPipelineConfig, accelerator: "Accelerator | None" = None):
 
     # create dataloader for offline training
     if not cfg.dataset.streaming:
-        # All non-streaming (map-style) datasets use EpisodeAwareSampler. This is broader than the
-        # historical `hasattr(active_cfg, "drop_n_last_frames")` guard: configs that previously fell
-        # back to DataLoader's default random shuffle now get this sampler instead, so their data
-        # order changes for a given seed (a deliberate, reproducibility-breaking improvement).
-        #
+        # All non-streaming (map-style) datasets use EpisodeAwareSampler.
         # The order is a pure function of (seed, epoch), so every rank independently produces the
         # same permutation. accelerate then shards it disjointly across ranks via BatchSamplerShard
         # without needing a `generator` attribute to synchronize an RNG, and resume is sample-exact.

tests/datasets/test_sampler.py

@@ -213,22 +213,6 @@ def test_deterministic_sampler_resume_is_exact_at_scale():
     assert list(resumed) == epoch_0[start:]
 
 
-def test_deterministic_sampler_validation_matches_episode_aware():
-    with pytest.raises(ValueError, match="drop_n_first_frames must be >= 0"):
-        EpisodeAwareSampler([0], [10], drop_n_first_frames=-1)
-    with pytest.raises(ValueError, match="drop_n_last_frames must be >= 0"):
-        EpisodeAwareSampler([0], [10], drop_n_last_frames=-1)
-    with pytest.raises(ValueError, match="No valid frames remain"):
-        EpisodeAwareSampler([0, 1, 2], [1, 2, 3], drop_n_first_frames=1)
-
-
-def test_deterministic_sampler_partial_episode_drop_warns(caplog):
-    with caplog.at_level(logging.WARNING, logger="lerobot.datasets.sampler"):
-        sampler = EpisodeAwareSampler([0, 1], [1, 6], drop_n_first_frames=1, shuffle=False)
-    assert list(sampler) == [2, 3, 4, 5]
-    assert "Episode 0" in caplog.text
-
-
 def test_compute_sampler_state():
     # 100 frames, batch 10, 2 ranks -> 10 underlying batches, 5 per rank per epoch.
     assert compute_sampler_state(step=0, num_frames=100, batch_size=10, num_processes=2) == {