docs: e-posta sınıflandırma benchmark eklendi (zero-shot %56, JSON %100)

docs: Türkçe karakterler düzeltildi — ç, ş, ğ, ı, ö, ü, İ

docs: bilingual TR/EN model card with ailm benchmark results (70/70 JSON, 1.2s/event)

Fix: update cost to $75, clarify model size

Initial release: Jazari-4B-SFT-TR (experimental Turkish adaptation of Qwen3.5-4B)

initial commit