fix: add plugin deploy script and fix Dockerfile plugin directory

Root cause of E2E failures: the E2E tests run against a live Headlamp
instance but never deploy the current PR's plugin code. The deployed
plugin still uses the old registerPluginSettings name, so the settings
component never renders.

- Fix Dockerfile output directory from headlamp-polaris-plugin to
  polaris, matching the deployed plugin identity
- Add scripts/deploy-plugin-to-headlamp.sh to build and deploy the
  plugin to Headlamp via kubectl before E2E tests

The e2e.yaml workflow needs a matching update to call this script
before running Playwright (requires workflows permission).

Co-Authored-By: Paperclip <noreply@paperclip.ing>

scripts/deploy-plugin-to-headlamp.sh

@@ -0,0 +1,70 @@
+#!/usr/bin/env bash
+# Deploy the built plugin to a live Headlamp instance via kubectl.
+#
+# Prerequisites:
+#   - kubectl configured with access to the Headlamp namespace
+#   - Plugin already built (npm run build → dist/)
+#
+# Environment variables (all optional, with defaults):
+#   HEADLAMP_URL        — Headlamp URL for readiness check
+#   HEADLAMP_NS         — Kubernetes namespace (default: kube-system)
+#   HEADLAMP_PLUGIN_DIR — Plugin path inside the pod (default: /headlamp/static-plugins/polaris)
+#
+# Usage:
+#   npm run build
+#   ./scripts/deploy-plugin-to-headlamp.sh
+#
+# Intended to be called from the E2E workflow before running Playwright tests,
+# so that E2E always tests the current commit's plugin code rather than whatever
+# was previously deployed.
+
+set -euo pipefail
+
+HEADLAMP_URL="${HEADLAMP_URL:-http://headlamp.kube-system.svc.cluster.local}"
+HEADLAMP_NS="${HEADLAMP_NS:-kube-system}"
+HEADLAMP_PLUGIN_DIR="${HEADLAMP_PLUGIN_DIR:-/headlamp/static-plugins/polaris}"
+
+if [ ! -d "dist" ]; then
+  echo "Error: dist/ not found. Run 'npm run build' first." >&2
+  exit 1
+fi
+
+# Find the Headlamp pod
+POD=$(kubectl get pod -n "$HEADLAMP_NS" -l app.kubernetes.io/name=headlamp \
+  -o jsonpath='{.items[0].metadata.name}')
+echo "Headlamp pod: $POD"
+
+# Remove stale plugin and copy current build
+kubectl exec -n "$HEADLAMP_NS" "$POD" -c headlamp -- \
+  rm -rf "$HEADLAMP_PLUGIN_DIR" 2>/dev/null || true
+kubectl cp dist/. "$HEADLAMP_NS/$POD:$HEADLAMP_PLUGIN_DIR" -c headlamp
+
+# Copy package.json so Headlamp can read plugin metadata
+kubectl cp package.json "$HEADLAMP_NS/$POD:$HEADLAMP_PLUGIN_DIR/package.json" -c headlamp
+
+# Verify the copy
+echo "Deployed files:"
+kubectl exec -n "$HEADLAMP_NS" "$POD" -c headlamp -- \
+  ls -la "$HEADLAMP_PLUGIN_DIR"
+
+# Restart the Headlamp process to reload plugins.
+# Killing PID 1 restarts the container without replacing the pod,
+# so the emptyDir volume (and our copied files) is preserved.
+echo "Restarting Headlamp process..."
+kubectl exec -n "$HEADLAMP_NS" "$POD" -c headlamp -- kill 1 || true
+
+# Wait for Headlamp to come back
+echo "Waiting for Headlamp to restart..."
+sleep 10
+for i in $(seq 1 30); do
+  HTTP_CODE=$(curl -s -o /dev/null -w '%{http_code}' --connect-timeout 5 "$HEADLAMP_URL" || true)
+  if [ "$HTTP_CODE" = "200" ]; then
+    echo "Headlamp is ready"
+    exit 0
+  fi
+  echo "  attempt $i/30 — HTTP $HTTP_CODE"
+  sleep 5
+done
+
+echo "Error: Headlamp did not recover after plugin deploy" >&2
+exit 1