Add model.tpack (browser payload for in-demo inference)

Promote seed1 weights to repo root so the Hub detects safetensors

Add NanoFable-28M-fp16 weights (fp16, seeds 0 and 1)

Track *.tpack with Git LFS

initial commit