fix: Wait for backend health check before starting frontend — prevents race condition during model loading

fix: Use python:3.10-slim base with Node binaries copied from node:20-slim via Docker COPY — zero apt for Node

fix: Restore full requirements.txt with CPU-only PyTorch for HF deployment

fix: Use lightweight requirements-deploy.txt (no PyTorch/transformers/whisper) to fit HF free tier size limits — ML services already have graceful fallbacks

fix: Complete Dockerfile rewrite — multi-stage build with Node 20 base, Python3 via apt, CPU-only PyTorch, and .dockerignore

fix: Add build-dependencies for PyTorch, Librosa, and bcrypt, and compel CPU-only PyTorch download to prevent Hugging Face Docker Container Out-Of-Memory crashes

fix: Simplify Docker Node.js installation to use native Debian packages to avoid environment symlink errors

fix: Replace Nodesource script with Docker multi-stage Node binaries injected to bypass build cache errors on Hugging Face

chore: Add Dockerfile, start.sh, and Next.js rewrites to support unified single-port deployment on Hugging Face Spaces