Optimize: Add --preload to gunicorn for stable startup

Fix: Backend OOM crashes via Vector Cache and worker reduction

Perf: Implement extensive loading time optimizations

Initial backend API deployment