Support optional dataset downloading from Hugging Face with caching

[flferretti]

This pull request introduces enhancements to the AMP loader by adding support for downloading datasets from Hugging Face using the login in https://github.com/ami-iit/amp-rsl-rl/blob/main/example/load_amp_dataset.py. The changes primarily focus on extending the AMPLoader class and its integration, along with minor updates to dependencies.

Enhancements to AMP functionality:

• amp_rsl_rl/runners/amp_on_policy_runner.py: Updated the initialization of AMPLoader to include new parameters such as download_from_hf, robot_folder, and repo_id, enabling dataset downloads from Hugging Face.
• amp_rsl_rl/utils/motion_loader.py: Modified the AMPLoader class to support optional parameters for downloading datasets from Hugging Face, including handling cache directories and repository details. Added a fallback to create a cache directory using platformdirs if dataset_path_root is not provided.

Dependency updates:

• pyproject.toml: Added platformdirs>=4.0.0 to dependencies for managing cache directories required by the new Hugging Face dataset download functionality.

Minor improvements:

• amp_rsl_rl/utils/motion_loader.py: Added Optional to type hints for better clarity and flexibility in function signatures.

[Giulero]

Thanks @flferretti! Had a look at it.
With this, is the dataset downloaded from ami hf, and just for ergoCub?
I'm a bit concerned about this, as ideally this library should also work for other robots!

[GiulioRomualdi]

I would like to keep the motion_loader independent from Huggingface. There should be a function that allows the user to download the dataset from HF. I think it makes the usage of the repo a bit cleaner.

[GiulioRomualdi]

I would like to keep the motion_loader independent from Huggingface. There should be a function that allows the user to download the dataset from HF. I think it makes the usage of the repo a bit cleaner.

The reason why of this is also because there could be several different place where the dataset is stored.