Try detecting sudo when running rustup-init

[inejge]

Re #608. Not all sudo setups keep HOME, e.g., the ones in RH-derived distros don't. For those, a sudo'd installation will not end up in user's home directory, which may be unexpected, but not ugly like rust-lang/cargo#2892. On Debian/Ubuntu and OS X the detection works as expected.

[inejge]

The Windows build failure is spurious.

Other failures are much more interesting.

1. Linux builds triggered a "can't get user data" warning inside the sudo check, and messed up expected test output. Unless /etc/passwd is nonexistent/unreadable, which is doubtful, the only other way it could've failed is if the euid under which the program is running didn't have a passwd entry. Not critical, but annoying.
2. OS X builds massively failed because the test harness modifies HOME, confusing the sudo check. If Linux builds haven't failed at (1), they would've hit this.

Since the sudo check seems fundamentally incompatible with the way in which the test harness works, I've added an override which skips it when testing. It's a kludge; maybe a #cfg would be cleaner?

[brson]

Looks great @inejge! I'd prefer this code be sunk into the self_update module, which has some other pre-flight checks. I want to do a release shortly though, so I'll do that refactoring myself. Hope you don't mind.