MultiDataset

I am using lerobot pi0 to train a multi-task policy. Will it work if i provide a list of repo_id and remove the exception part of make_dataset.

Start by reading the make_dataset entry point and the exception mentioned in the issue. Trace how repo_id is currently accepted and determine whether a list can support pi0 multi-task training without breaking existing behavior. Done means the supported input and resulting dataset behavior are defined and verified.

Written by the indexing model from the issue text.