The DeLong variance in the AUC gaps explainer is quadratic and takes 17 seconds #1169

Where: explainers/auc-gaps-by-group.md (delong_auc_var), explainers/code/auc-gaps-by-group.py.

Current gap: The placement values are computed with a Python loop over every positive and every negative ((neg < p).mean()), so a group of 10,000 rows takes most of the script's 17 seconds. The other five new explainer scripts take under 2.5 seconds.

Repro:

$ time PYTHONPATH=. python3 explainers/code/auc-gaps-by-group.py   # real 17s

Suggested approach: Compute the placement values from ranks (scipy.stats.rankdata) in O(n log n), check the numbers are unchanged, and update the markdown.

Done when: the script runs in under 3 seconds with identical printed output.