Make e2e workflow a proper on-demand regression harness

- workflow_dispatch input to filter by test name (-PtestNames)
- 25-min job timeout
- cache Paper jar + libraries + BentoBox/BSkyBlock deps + npm modules
- always upload server log + plugwright report as an artifact (so a
  regression failure is debuggable)

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NKxodNE4h3TsSHMqDEeC8v

Author: tastybento

.github/workflows/e2e.yml

@@ -1,16 +1,28 @@
 name: E2E (plugwright)
 
-# In-game end-to-end tests via plugwright (boots a real Paper server + Mineflayer bot).
-# Advisory / opt-in for now: run manually from the Actions tab. It is intentionally NOT wired
-# to every push/PR yet (a full run boots a server + generates worlds, ~1-2 min, and is heavier
-# and flakier than the Maven unit tests). Add a `pull_request` trigger once it's proven stable.
+# In-game end-to-end tests via plugwright (boots a real Paper server + Mineflayer bot) —
+# the layer the Maven/MockBukkit unit tests can't reach. Run ON DEMAND as a regression check.
+#
+# Trigger it from the GitHub UI (Actions tab -> "E2E (plugwright)" -> "Run workflow", pick the
+# branch/tag) or from the CLI:  gh workflow run e2e.yml --ref <branch-or-tag>
+#
+# It is deliberately NOT wired to every push/PR: a run boots a server and generates worlds
+# (~1-2 min) and is heavier/flakier than the unit tests. For a nightly regression, add:
+#   schedule:
+#     - cron: "0 3 * * *"
 on:
   workflow_dispatch:
+    inputs:
+      test_names:
+        description: "Only run tests whose name contains this (comma-separated substrings). Blank = all."
+        required: false
+        default: ""
 
 jobs:
   e2e:
     name: E2E tests
     runs-on: ubuntu-latest
+    timeout-minutes: 25
     steps:
       - uses: actions/checkout@v4
 
@@ -25,9 +37,50 @@ jobs:
         with:
           node-version: 20
 
+      - name: Set up Gradle (with caching)
+        uses: gradle/actions/setup-gradle@v4
+
+      # Reuse the Paper jar + downloaded libraries and the BentoBox/BSkyBlock jars across runs.
+      # plugwright's clean step preserves server.jar / cache / libraries, and build.gradle.kts
+      # caches the dependency jars in .deps, so this avoids re-downloading ~60 MB every run.
+      - name: Cache Paper + dependency jars
+        uses: actions/cache@v4
+        with:
+          path: |
+            e2e/.deps
+            e2e/run/server.jar
+            e2e/run/cache
+            e2e/run/libraries
+          key: plugwright-${{ hashFiles('e2e/build.gradle.kts') }}
+
+      - name: Cache npm modules
+        uses: actions/cache@v4
+        with:
+          path: e2e/src/test/e2e/node_modules
+          key: e2e-npm-${{ hashFiles('e2e/src/test/e2e/package.json') }}
+
       - name: Build Challenges jar (Maven)
         run: mvn -B -DskipTests package
 
       - name: Run E2E tests
         working-directory: e2e
-        run: ./gradlew plugwrightTest --no-daemon
+        env:
+          PLUGWRIGHT_DEBUG: "1"
+        run: |
+          ARGS=""
+          if [ -n "${{ inputs.test_names }}" ]; then
+            ARGS="-PtestNames=${{ inputs.test_names }}"
+          fi
+          ./gradlew plugwrightTest $ARGS --no-daemon
+
+      # Always upload the server log + any plugwright report so a regression is debuggable.
+      - name: Upload server log & report
+        if: always()
+        uses: actions/upload-artifact@v4
+        with:
+          name: e2e-logs
+          path: |
+            e2e/run/logs/**
+            e2e/run/plugwright-report/**
+          if-no-files-found: ignore
+          retention-days: 14