chore: fix updates command

site/update-content.sh

@@ -1 +1,26 @@
- nu ../../../website-htmx-rustelo/code/provisioning/content.nu sync --root "$(pwd)" # → snapshot al PV (/var/www/htmx-assets). Ya sin --all.
+#!/usr/bin/env bash
+# Publish this site's tree to the running pod, letting content.nu choose the
+# MINIMAL correct action from the changeset's reload class (distro.ncl ::
+# reload_classes): hot → no restart · restart → snapshot + rolling restart ·
+# rebuild → refused (that one needs a new image, not content).
+#
+#   ./update-content.sh                 # classify + publish
+#   ./update-content.sh --dry-run       # say what it would do, touch nothing
+#   ./update-content.sh --namespace x   # any content.nu flag passes through
+#
+# Deploy identity (namespace / app_label / pvc / mount / cp_node) is resolved
+# from rustelo.manifest.toml — no literals here.
+#
+# NOT `sync`: sync packs only the `hot` subset and never restarts, so a change
+# under site/config or site/i18n/locales shipped green while the pod kept
+# serving the old one.
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+CONTENT_NU="${CONTENT_NU:-$SCRIPT_DIR/../../../website-htmx-rustelo/code/provisioning/content.nu}"
+
+[ -f "$CONTENT_NU" ] || { echo "content.nu not found: $CONTENT_NU" >&2; exit 1; }
+
+# --since is inert while this tree is not its own git root (it lives under the
+# outreach repo), so content.nu falls back to a full snapshot on every run.
+exec nu "$CONTENT_NU" publish --root "$SCRIPT_DIR" "$@"