Failing tests in test.ml, fork.ml and trace.ml:

• test.ml failed because it expects memtrace to catch and report an exception only once. Thus, in the lock_tracer function we need to check for failure before trying to obtain the lock. Otherwise, the tracer will continue to attempt to write traces even after it has failed, and will likely throw the same error multiple times. After switching the order of "if s.failed" and "if Mutex.try_lock", the test passes.
• another issue was that a thread that already holds the lock may try to obtain it again because (1) an allocation happens while a thread is in the "close tracer" function or (2) if a thread is writing an external allocation. If we replace Mutex.try_lock with Mutex.lock, it will throw an error if this happens. we catch this error and return false to avoid deadlocks.