Add eu-estimate.py + eu-lang-speakers.json: estimate(cc)=K*Σ_lang authors(lang)* speakerWeight(cc|lang), K calibrated so NL matches its own model. Merges an `estimate` per country + `peopleEstimate` into europe-stats.json. HomePage hero now leads with the estimate for be/eu/no (like nl) and falls back to the attributed count otherwise. English- first gb/ie are excluded (English untracked); es/pt under-covered. Wired into the NAS job after eu-stats-aggregate; language is estimate-layer only, never confirms a country.