fix(helm): hardening from review

- _helpers.tpl: required guard on image.tag — silent empty deploys can
  no longer happen; helm fails fast with a clear message.
- configmap-nginx: HTTPS fastcgi param uses if_not_empty, so PHP only
  sees HTTPS when X-Forwarded-Proto is actually present.
- deployment-app: add startupProbe with 100s budget so first-boot cache
  warmup doesn't trip liveness.
- deployment-horizon: failureThreshold=5 on the horizon:status probe;
  transient Valkey blips no longer cause restart loops.
- job-migrate: mount oauth-keys so seeders that touch Passport clients
  don't silently fail.
- statefulset-valkey: replace separate password Secret with a
  requirePassword toggle that reads REDIS_PASSWORD from the main chart
  Secret (same Secret app/horizon/scheduler already mount). Liveness
  probe authenticates with the password when set.
- values-us-prod: enable valkey.requirePassword.
- README: add REDIS_PASSWORD to bootstrap procedure.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

helm/ezscale-website/values.yaml

@@ -77,7 +77,11 @@ mariadb:
 valkey:
   enabled: true
   image: valkey/valkey:9-alpine
-  password: ""              # if empty, chart generates a random secret
+  # When true, valkey runs with --requirepass and reads REDIS_PASSWORD from
+  # the main chart Secret (same key the app/horizon/scheduler pods read via
+  # envFrom). The Secret MUST contain a REDIS_PASSWORD key — bootstrap it
+  # alongside APP_KEY in production, or include it in secret.values for dev.
+  requirePassword: false
   maxmemory: "1gb"
   storage:
     size: 5Gi