This was first noticed while testing with async-profiler https://github.com/async-profiler/async-profiler/issues/1617

During VMDeath, the flight recorder will be shutdown, which in turn causes the `recorderthread` to complete, thus the flight recorder will no longer be able to process any new recordings

It's possible for a user application to call `jdk.jfr.Recording#start` after the `recorderthread` is completed, which means that this new recording will be stuck infinitely waiting for the `recorderthread` to process the `MSG_START` which will never happen as the thread is already dead.

If that happen, then that thread will forever be stuck & potentially causing the application to deadlock & fail to die

This was reproduced on all JDK8+ including tip on both Linux & macOS

The following small program should reliably reproduce this deadlock

```
import jdk.jfr.Recording;

public class JfrHang {

    public static void main(String[] args) {
        // Initialize JFR subsystem
        Recording r = new Recording();
        r.start();
        r.stop();

        Runtime.getRuntime().addShutdownHook(new Thread(() -> {
            System.out.println("Starting recording");
            try {
                Thread.sleep(100);
            } catch (Exception e) {}
            new Recording().start();
            System.out.println("Recording started");
        }));

        System.out.println("Exiting");
    }
}
```