convert_aloha_data_to_lerobot.py is outdated and extremely memory-heavy

[AaronDsouza-2006]

The current convert_aloha_data_to_lerobot.py script has two major issues that makes it difficult to train openpi:

1. Outdated LeRobot format (v2 instead of v3.0)

   • The script produces the old LeRobot v2 dataset format.

2. Extremely high memory usage

   • The script loads too much data into RAM at once
   • It does not support proper batching

Steps to Reproduce

1. Prepare a folder with ALOHA-style episode_*.hdf5 files
2. Run convert_aloha_data_to_lerobot.py.
3. After conversion, the resulting dataset is in the old v2 format.

Expected Behavior

• The script should output a LeRobot v3.0 compatible dataset by default.
• It should support batching / low-memory mode (process episodes in small groups, compute running statistics, write Parquet chunks incrementally).