ExtremeAIGC: Benchmarking LMM Vulnerability to AI-Generated Extremist Content