fix(installer): align installer/onboarding with current codebase (Phase 1 audit)

A full audit of the installer and first-run experience found the mandatory
first-run wizard (is_initial_setup gate + InitialSetup.tsx) already works
correctly end-to-end — the real gap was discoverability and drift between
the installer and current app, not the wizard itself:

- nginx's static welcome page (served on the default HTTP/HTTPS ports, the
  first place any new admin looks) never mentioned the admin panel lives on
  a separate port, still shipped a Korean-language switcher despite Korean
  being fully removed from the product, and linked to the retired
  "Argus-appliance" identity. Now surfaces a live, dynamically-derived
  admin panel link and drops the dead i18n/branding.
- users.language defaulted to 'ko' at the schema level (both the column
  DEFAULT and the seed admin row) — a leftover from before Korean removal,
  silently affecting every new user, not just the seed account.
- Three independently-drifting install paths existed (public distribution
  installer, private-repo SSH-ship installer, and a fully manual README/docs
  flow). README/docs now lead with the public alleyviper/argus installer
  (verified to actually exist and mirror distribution/); the manual path is
  kept only as an explicit private/pre-release fallback.
- Both scripted installers auto-rotated admin/admin via the API as their
  last step, silently warning (not failing) if it didn't work — inconsistent
  with the manual path and capable of leaving defaults active undetected.
  Removed; every install path now consistently lands on admin/admin + the
  in-app wizard.
- ANIS_ADMIN_KEY was documented as required ("ARGUS fails closed without
  it") but is never read anywhere in src/api — confirmed dead, not just
  vestigial. Retired from env templates, compose files, and docs; ROADMAP's
  existing "wire it up or retire it" item resolved for this variable.
- Proxy host creation returned an opaque 500 when given an unresolvable
  forward_host (e.g. a Docker container name typed into the free-text
  field — ARGUS's nginx runs in host-network mode with no embedded DNS).
  nginx -t's real "host not found in upstream" failure now surfaces as an
  actionable 400 instead of a generic internal error; placeholder text
  fixed to stop inviting the failure. Live-verified both the failure path
  (with NGINX_SKIP_TEST temporarily disabled in the sandbox to exercise the
  real validation) and the no-regression path (valid IP still succeeds).
- LOG_COLLECTION silently disables the entire access-log/analytics/live-event
  pipeline when set to anything but the exact string "true", with zero log
  output anywhere. Added a boot-time WARN matching the existing
  IsInitialSetupRequired nudge pattern.

Every fix live-verified against a genuinely fresh sandbox install
(docker/docker-compose.local-sandbox.yml, zero pre-existing volumes),
not just read from source.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

docker-compose.yml

@@ -123,7 +123,6 @@ services:
       ANIS_ENABLED:          ${ANIS_ENABLED:-false}
       ANIS_URL:              ${ANIS_URL:-}
       ANIS_LICENSE_KEY:      ${ANIS_LICENSE_KEY:-}
-      ANIS_ADMIN_KEY:        ${ANIS_ADMIN_KEY:-}
       ANIS_SHARE_ATTACKERS:  ${ANIS_SHARE_ATTACKERS:-false}
       DNS_SECURITY_LISTEN_ADDR: ":53"
     ports:
@@ -196,6 +195,7 @@ services:
         max-file: "5"
     environment:
       TZ: ${TZ:-UTC}
+      UI_PORT: ${UI_PORT:-81}
     ulimits:
       nofile:
         soft: 65535