EgoCross: Benchmarking Multimodal Large Language Models for Cross-Domain Egocentric Video Question Answering