Skip hidden .parquet files when loading dataset to avoid PyArrow errors

[ahadjawaid]

What this does

Fixes handling of hidden .parquet files during dataset loading (Bug).
Some environments generate hidden files like ._file-000.parquet, which cause Dataset.from_parquet to error when memory-mapped loading is used.
This PR filters out such hidden .parquet files before loading.

How it was tested

• Reproduced the issue with paths containing both hidden and regular .parquet files.
• Verified that dataset loading succeeds after filtering out hidden files.
• Confirmed that normal dataset loading remains unchanged when no hidden files are present.

How to checkout & try? (for the reviewer)

python -c "from lerobot.datasets.utils import load_dataset; load_dataset('/path/to/episodes')"

or run any pipeline relying on Dataset.from_parquet to confirm it loads without errors when hidden files exist.