For i386 temporary mappings, unpin the thread before releasing
the cmap lock.  Releasing the lock first may result in the thread
being immediately rescheduled and bound to the same CPU, only to
unpin itself upon resuming execution.

Noted by:	skra (in review for armv6 equivalent)

Add comment explaining relative order of sched_unpin() and mtx_unlock().

Suggested by:	alc