Update model card: argmax + PhilBerta ensemble, current alignment counts

Update usage examples from itermax to argmax

Update model card for v3: drama training data, 15 MG passes, current pipeline numbers

Update tokenizer_config.json with v3 (AG-MG theatre) checkpoint

Update tokenizer.json with v3 (AG-MG theatre) checkpoint

Update special_tokens_map.json with v3 (AG-MG theatre) checkpoint

Update model.safetensors with v3 (AG-MG theatre) checkpoint

Update config.json with v3 (AG-MG theatre) checkpoint

Update Dilemma form count (12.5M) and AG Classical accuracy (99.7%)

Update Dilemma accuracy: AG Classical 96.9%, Katharevousa 95.6%

Add test suite for model card, config, tokenizer, weights, and inference

Document prev_word and dialect parameters for Dilemma integration

Update model card: LSJLogeion -> lsj9, sync Dilemma accuracy numbers

Update model card: Dilemma 12.3M forms, current benchmarks, Opla, fix training time and AG-MG numbers

Update model card: AER results, prose + NT training data, 128K alignments

Add YAML metadata to model card (CC-BY-4.0, tags, languages)

Dragoman: multilingual Greek word alignment (AG-EN, AG-MG)