DEPLOY_ENV=production now marks the primary instance globally - used for
Caddy ownership, the Dev/Prod header badge, and Caddy UI gating. Removes
build-time VITE_DEPLOY_ENV/import.meta.env.DEV from the header in favour
of the runtime API response (isProduction field in /api/auth/config).
Both instances run with NODE_ENV=production, so import.meta.env.PROD was
always true and the header always showed 'Production'. deploy.sh now passes
VITE_DEPLOY_ENV=<branch> into the build and Header reads it to label the
system indicator dev vs prod correctly.
One Caddy serves the whole container and POST /load replaces the entire
config, so two instances pushing would clobber each other. Now only the
instance with CADDY_MANAGER=true (production) pushes, seeds routes from the
Caddyfile, and accepts route mutations (others get 403). /api/auth/config
exposes caddyManaged so the non-owner Settings UI shows the Caddy section
read-only. The installer sets the flag on the production .env only.
On every startup, if Caddy is already enabled but the caddy table has no
routes (e.g. after a re-deploy), importCaddyfileRoutes() is called so the
static GhostGrid entries from /etc/caddy/Caddyfile are seeded automatically.
Also ensures deploy.sh is executable via proxmox-ghostgrid.sh.
