In our forked DNS resolver processes, don't bail out if we get a retryable error while reading from our parent socket. I think EAGAIN and EWOULDBLOCK never happen because this is a non-blocking socket AND because we're using select (seems like we'd be fine using one or the other, and it's not necessary to use both?). I believe this is correct behavior and we should be doing this same thing everywhere we read from sockets. I think we probably don't check for EINTR in many places where we should.