fix(memory): do not close cached manager during doctor status probe

The doctor.memory.status handler was closing the manager in a finally
block, but MemoryIndexManager.get returns a cached instance that may
be the long-lived startup manager with an active file watcher. Closing
it tears down the watcher and disables external-change auto-reindexing.

Remove the close call — manager lifecycle is handled by
closeAllMemorySearchManagers at gateway shutdown.

src/gateway/server-methods/doctor.ts

@@ -56,11 +56,9 @@ export const doctorHandlers: GatewayRequestHandlers = {
       };
       respond(true, payload, undefined);
     } finally {
-      try {
-        await manager.close();
-      } catch {
-        // Ignore close errors - response already sent
-      }
+      // Do NOT close the manager here — it may be the long-lived cached
+      // instance from startup whose file watcher must stay active.
+      // Manager lifecycle is handled by closeAllMemorySearchManagers at shutdown.
     }
   },
 };