Academic-style training methodology documentation

Optimize: Remove training state, keep only model weights (5.4GB -> 1.9GB)

Add evaluation results from training

Fix limitations: clarify reference audio not needed for voice cloning after fine-tuning

Fix config.json: replace Infinity with 1e9 for valid JSON

Remove "The first" from description

Upload XTTS v2 Antonio Oliveira PT-PT fine-tuned model

initial commit