Data and Benchmark for VideoLoom: A Video Large Language Model for Joint Spatial-Temporal Understanding