v1.5.0 — Stability, performance & polish Highlights: - CUDA VRAM decode (-hwaccel_output_format cuda) — prevents OOM on low-memory hosts - SIGKILL retry limits — max 3 retries with exponential backoff - Boot order fix — app.listen() before encoder.start() for fail-fast on port conflict - Single process manager — installer picks systemd or PM2, never both - GPU device isolation — CUDA_VISIBLE_DEVICES + hwaccel_device 0 - EADDRINUSE graceful handling - Refactored encoder into ffprobe.js, ffmpeg-args.js, webhook.js - Fixed all 42 empty catch blocks - Full code quality audit - Complete README rewrite - Polished install.sh, Dockerfile, docker-compose See CHANGELOG in README.md for full details.