RAG vs. Long Context: Examining Frontier Large Language Models for Environmental Review Document Comprehension