docs: update training data distribution with accurate numbers (SFT 36,713 + DPO 24,779)

docs: update model card with GGUF formats, benchmarks, usage examples

Fix year to 2026

Update model card for DPO v4

initial commit