OBLIQ Benchmark Emerges As Key Test For AI Agent Observability