eval: add eval_results.json, model-index, fix evaluate.py NaN→null

docs: clean benchmark section + proper model-index (mean metrics, no per-dataset wall)

eval: add model-index eval results + benchmark section to README

Add WEO-SAS model card