Reduce FPM workers from 12 to 5 max to stop memory thrashing on
a single-core VPS with 3GB RAM. Add OPcache and APCu tuning to
reduce filesystem stat calls and improve cache hit rates.
Increase PHP-FPM max_children from 5 to 12 to handle concurrent
requests without queuing, sized for a ~3GB VPS. Add Mermaid
architecture diagram to README.
Replace the all-in-one Apache image with nextcloud:33-fpm and an Nginx
sidecar that serves static assets directly with gzip compression and
cache headers, avoiding the prefork concurrency bottleneck.
