libnative/io/file_unix: remove superfluous retry().

[nodakai]

Those syscalls listed below don't return EINTR, so wrapping them with retry() is superfluous.

But I admit the current code is better from the viewpoint of difensive programming, given that the overhead of retry() is really cheap...

http://pubs.opengroup.org/onlinepubs/9699919799/functions/fstat.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/mkdir.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/unlink.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/rename.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/rmdir.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/readlink.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/symlink.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/link.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/fstatat.html
http://pubs.opengroup.org/onlinepubs/9699919799/functions/utime.html

[lilyball]

I think it's probably a good idea to remove retry() around calls that can't return EINTR. At the very least, it avoids confusion when the reader sees retry() and believes that means the call can in fact return EINTR. @alexcrichton?

[alexcrichton]

I'm a little wary with this in that are we 100% certain that every single implementation of these syscalls will never return EINTR? I certainly agree that according to the documentation in the linked manpages none of these can return EINTR, however!

Regardless, it's pretty easy to add these all back at a later date, so I'm ok removing them for now because in theory any compliant implementation won't return EINTR.

[lilyball]

@alexcrichton I was wary too, but I checked a random sampling of the functions in the manpages for both OS X and Linux, and every function I looked at matched the specification. I would assume a non-compliant implementation would cause errors in lots of programs, so I think it's safe to assume that won't happen.