Contribution description

This PR fixes a rare race-condition in xtimer_msg_receive_timeout which can lead to corruption of the timer list and subsequent hard faults.
The race condition is triggered when:

1. A message is sent to a thread using msg_send.
2. The xtimer which sents the timeout message expires and executes before xtimer_remove(t) is called. This will cause the message queue for the thread to contain first a real message and second the timeout message. The timer will not be queued anymore, but the timeout message will still be in the queue.
3. The xtimer_msg_receive_timeout function is called again. This will queue a new xtimer while the timeout message of the previous timer is still in the buffer. _msg_wait will see the old timeout message, think that the current xtimer has already expired and will not remove the timer. When xtimer_msg_receive_timeout returns, the timer will still be queued. However, as it is allocated on the stack it is no longer valid. This causes the timer_list_head to now point to invalid memory. Crashes ensue.

Testing procedure

This bug was found, validated, and fixed using a proprietary application. I have not written a separate example application which exhibits the problem which I could publish.