Predictive performance of presence‐only species distribution models: a benchmark study with reproducible code