Fix graceful shutdown when intermediate or init process errors or panic

[yihuaf]

Fix #237 Tracking #221

Gracefully shutdown by closing down unused fds in the channel. In this case, when the child process exits, the parent waiting (read_exact) is exit blocking and error out. When fork, the fds in the channel are duplicated, so there are two receivers and two senders. Since we force the channel to become uni-directional, we can safely close the additional fds.

Also in the PR, we should only set host name when entering into a new UTS namespace. One of the integration test doesn't create any new uts namespace, so calling set_hostname is not a good idea.

Now pass one more integration test. The test essentially asks Youki to enter into the wrong type of namespace and see if Youki will error out.

[utam0k]

How about this code?

Suggested change

	unistd::close(self.receiver.as_raw_fd())?;
	Ok(())
	}
	pub fn close_sender(&self) -> Result<()> {
	unistd::close(self.sender.as_raw_fd())?;
	Ok(())
	unistd::close(self.receiver.as_raw_fd())
	}
	pub fn close_sender(&self) -> Result<()> {
	unistd::close(self.sender.as_raw_fd())

[yihuaf]

unistd functions will return a different error type than anyhow result, so we have to use ? for a conversion.