REFACTOR GAIA should use get_dataset instead of direct call to huggingface_hub

[juanmichelini]

GAIA currently uses

load_dataset("gaia-benchmark/GAIA", level) - direct call, NOT going through get_dataset() from dataset.py
huggingface_hub.snapshot_download(...) - direct call

It should use get_dataset() like other benchmarks instead, to reduce code duplication