Fix TypeError in LeRobotDataset initialization when using --control.resume=true flag

[cdborinstein]

What this does

Fixes TypeError in LeRobotDataset initialization when using --control.resume=true flag. (🐛 Bug)
The current code incorrectly uses torch.stack() on single tensor objects in lines 560-561 of /lerobot/common/datasets/lerobot_dataset.py, causing:

TypeError: stack(): argument 'tensors' (position 1) must be tuple of Tensors, not Tensor
TypeError: stack(): argument 'tensors' (position 1) must be tuple of Tensors, not Column

Root cause: torch.stack() requires a sequence of tensors as input, but the code was passing individual tensors/columns.
Solution: Removed unnecessary torch.stack() calls since the data is already in the correct tensor format:

Before:

pythontimestamps = torch.stack(torch.tensor(self.hf_dataset["timestamp"])).numpy()
episode_indices = torch.stack(self.hf_dataset["episode_index"]).numpy()

After:

pythontimestamps = torch.tensor(self.hf_dataset["timestamp"]).numpy()
episode_indices = torch.tensor(self.hf_dataset["episode_index"]).numpy()

How it was tested

• Reproduced the original error using the exact command from the documentation. Confirmed that no error occurs without --control.resume flag
• Applied the fix and verified it resolves the TypeError
• Tested complete data collection workflow with --control.resume=true flag
• Confirmed successful dataset upload to Hugging Face repository
• Verified no regression in normal recording functionality (without resume flag)

Test environment: Trossen Stationary AI running lerobot via HuggingFace.

How to checkout & try? (for the reviewer)

1. Reproduce the original error (on code before this fix):

python lerobot/scripts/control_robot.py \
--robot.type=trossen_ai_stationary \
--robot.max_relative_target=null \
--control.type=record \
--control.fps=30 \
--control.single_task="Test recording episode using Trossen AI Stationary." \
--control.repo_id=${HF_USER}/test_dataset \
--control.tags='["tutorial"]' \
--control.warmup_time_s=5 \
--control.episode_time_s=30 \
--control.reset_time_s=30 \
--control.num_episodes=2 \
--control.push_to_hub=true \
--control.resume=true

2. Verify the fix works (with this PR's changes):

# Same command as above - should now run without TypeError

3. Check the specific fix:

# Verify the problematic lines are fixed
grep -n -A 1 -B 1 "torch.tensor.*timestamp" lerobot/common/datasets/lerobot_dataset.py