Add root parameter to training script. Use --dataset.root_dir to set root dir

[rkeshwani]

Resolves issue #689.

What this does

Adds a root_dir parameter to the datasetConfig class which determines which parameters are accepted in the train.py script.

The make_dataset factory configuration was modified as well to actually use the parameter.

How it was tested

Data was recorded with a different root path.
Then the training was run with the new root dir specified and the training succesfully started. Before this implementation an error was thrown showing the user's home directory being used to search for the dataset.

Examples:

python train.py --dataset.repo_id=lerobot/test --policy.type=act --output_dir=./training_data --job_name=act_so100_test --device=cuda --dataset.local_files_only=true --dataset.root_dir=./recorded_datasets

Notes

-I'm unsure if the changes I made will impact the library in other areas like recording and evaluation because I changed the dataclass of the datasetconfig. If I find any issues I will submit as part of a separate pull request with the assumption this one is accepted.

[aliberts]

Thanks! I will close this as we just merged a duplicate of this: #765 (sorry I didn't see yours before).
You should now be able to do this with --dataset.root