Rewrite unwind thread handling.

This new version doesn't require any specialized thread
implementation, it uses the Current implementation to do its job.
In addition, it runs much faster when multiple threads are trying
to unwind at the same time since the global signal lock is held for
only a small amount of time. Even running through the threads one at
a time should be faster since it no longer requires two passes through
the unwound stacks.

The new code now allows multiple simultaneous unwinds of the
same thread.

Finally, add the ability to unwind from a ucontext_t passed in. This
functionality doesn't work for remote unwinds yet.

Change-Id: I4d181d7ca5ffd2acfd1686e668e6d21e36b425cb