Remove the unrelated TACRED-style test split and the generic duplicate split files introduced in the recent conversion commits. Restore the GENIA event-extraction dataset card and keep only the correctly named labeled train/validation files. The official blind test inputs have no public event gold labels, so no labeled test split is published.
