Sync KV-cached model.py to backend models directory

Implement KV caching for O(N) generation and speedups

Convert absolute module imports to relative to support Docker environments

Deploy: Prepare codebase for Hugging Face Spaces