fix: detect and auto-restart zombie gateway (executor shutdown)

The gateway can enter a zombie state where the process is alive and
health checks pass, but the asyncio executor has been shut down.
All message sends fail with 'RuntimeError: Executor shutdown has been called'.

The watchdog now scans the last 50 log lines for 'Executor shutdown'
errors. If detected, it force-kills and restarts the gateway.

entry.py

@@ -796,11 +796,40 @@ def _gateway_watchdog(interval: int = 30):
 
             gw_pid = _find_gateway_pid()
             if gw_pid is not None:
-                # Gateway is alive, reset backoff
-                restart_backoff = 30
-                continue
+                # Process is alive — but check if gateway is FUNCTIONAL (not zombie)
+                # A zombie gateway has a live process but dead asyncio executor,
+                # causing "Executor shutdown has been called" on every send.
+                try:
+                    # Check gateway log for executor shutdown errors in last 60s
+                    now_ts = time.time()
+                    _zombie = False
+                    if os.path.isfile(gw_log):
+                        with open(gw_log, "r", errors="replace") as _lf:
+                            for _line in _lf.readlines()[-50:]:
+                                if "Executor shutdown" in _line and "RuntimeError" in _line:
+                                    _zombie = True
+                                    break
+                    if _zombie:
+                        logger.error("[Watchdog] Gateway is ZOMBIE (executor shutdown) — force-restarting...")
+                        killed = _kill_all_gateways()
+                        if not killed:
+                            try:
+                                os.kill(gw_pid, 9)
+                            except Exception:
+                                pass
+                        time.sleep(3)
+                        # Fall through to restart below
+                    else:
+                        restart_backoff = 30
+                        continue
+                except Exception:
+                    # If log check fails, assume gateway is fine
+                    restart_backoff = 30
+                    continue
+            else:
+                logger.warning("[Watchdog] Gateway process is DEAD — restarting...")
 
-            # Gateway is dead — restart with backoff
+            # Gateway is dead or zombie — restart with backoff
             now = time.time()
             if now - last_restart < restart_backoff:
                 continue