feat: update parse_separated_files to resolve absolute paths for separated audio files

refactor: update model initialization to use explicit filename and improved logging

feat: implement FastAPI-based audio separation service with ZIP and stem-specific export endpoints for Hugging Face Spaces

initial commit