testing: prevent test scheduling after reactor exit

Previously, when handling "--help" (introduced in b8a13be), we returned
status code 0 but continued scheduling test tasks to the Seastar reactor.
This caused test applications to hang since the tasks expected the exchanger
'e' to be available after reactor exit.

Fix this by using the "_done" flag to track reactor state:
- Set "_done" when Seastar application exits
- Skip task scheduling and exchanger wait if "_done" is set
- Reuse existing "_done" member as it's safe in single-threaded context

This fixes the regression where "--help" would cause test applications to
hang indefinitely.

Fixes #2635
Signed-off-by: Kefu Chai <[email protected]>

src/testing/test_runner.cc

@@ -107,6 +107,7 @@ int test_runner::start_thread(int ac, char** av) {
                 return 0;
             });
         });
+        _done = true;
         init_outcome->give(_exit_code);
     });
 
@@ -130,6 +131,11 @@ test_runner::run_sync(std::function<future<>()> task) {
         // it.
         return;
     }
+    if (_done) {
+        // if "--help" is passed to the seastar app in the command line, _done is
+        // set before the seastar worker thread takes any task.
+        return;
+    }
 
     exchanger<std::exception_ptr> e;
     _task.give([task = std::move(task), &e] {