fix: restore None guard for legacy direct-stage callers

Tests that construct ZephyrCoordinator without initialize() leave
_coordinator_thread as None. The guard is needed for that path.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: yoblin

lib/zephyr/src/zephyr/execution.py

@@ -870,7 +870,7 @@ def _wait_for_stage(self) -> None:
 
             # Checked after completion so a clean shutdown racing the final
             # task can never false-positive — only true crashes reach here.
-            if not self._coordinator_thread.is_alive():
+            if self._coordinator_thread is not None and not self._coordinator_thread.is_alive():
                 raise ZephyrWorkerError(
                     "Coordinator thread is no longer alive. "
                     "Check logs for 'Coordinator loop crashed' for the root cause."