- Fixed the queue memory leak - Created a critical section for the queue - I removed the allocator lock when it calls NotifyRelease, I didn't add a case for being able to call NotifyRelease inside the lock because NotifyRelease never receives a pointer to the allocator so I don't think there is a need to still lock while running NotifyRelease. - Moved process_input to NotifyRelease. - Also removed the test because it doesn't reproduce the bug reliably specially in the testbot, but I can add it again if it's needed or try to look for a way to make it reliable.