app.json: drop redundant top-level patch_size (now read from Prediction.yml)

Update model card, app.json (patch size + VRAM plan) and inference configs

Update app.conf

Add finetune models

Add finetune models