Add BlockHostUntilDoneWithStatus, which returns Status rather than bool.

Also fixed a deadlock in Stream::BlockHostUntilDone.  The problem with the
original code was that it grabbed mu_ before looping over substreams, and would
call CheckError with mu_ still held.  But CheckError will attempt to lock mu_ in
the failure case, which would deadlock.

PiperOrigin-RevId: 178191634