debug(ci): tmate ssh into macos-14 node24 e2e job on failure

Cannot reproduce the macos-14 + node 24 watch e2e failure locally
on the same Node version (24.15.0), same fixture, same binding,
same concurrency. Last unknown is the GitHub Actions macos-14
runner's actual hardware/OS state. Open a tmate session on
failure, limited to that one matrix slot, so we can inspect:

  - runner CPU/memory pressure
  - the fresh-dir contents after fs.update
  - whether rspack input fs cache holds the old content
  - whether vnode invalidation lag is observable

Will be reverted after the investigation.

Author: stormslowly

.github/workflows/test.yml

@@ -226,6 +226,18 @@ jobs:
       - name: E2E Test (${{ matrix.test_script }})
         run: cd e2e && pnpm ${{ matrix.test_script }}
 
+      # RCA debug: SSH into the macos-14 + node 24 runner on e2e failure
+      # so we can inspect why the watch test sees stale dist content
+      # (mode B "No test files need re-run") that we cannot reproduce
+      # locally on the same Node version. Limited to the one matrix slot
+      # that consistently fails, so other failures are not impacted.
+      - name: Setup tmate session (RCA debug)
+        if: failure() && matrix.os == 'macos-14' && matrix.node_version == '24'
+        uses: mxschmitt/action-tmate@172994372753cb896adac383df48b1d7a2dae4d1 # v3.23
+        timeout-minutes: 30
+        with:
+          limit-access-to-actor: true
+
       - name: VS Code Extension Test
         if: matrix.test_script == 'test'
         run: pnpm run test:vscode