feature-gate most Nix functions

[asomers]

Using features reduces build time and size for consumer crates. By
default all features are enabled.

[asomers]

Supersedes #1498 . I'm creating a new PR rather than update the previous one because I didn't want to erase some of its history by force-pushing.

[rtzoeller]

bors r+

[bors]

Build succeeded:

• Android aarch64
• Android arm
• Android armv7
• Android i686
• Android x86_64
• DragonFly BSD x86_64
• FreeBSD amd64 & i686
• Fuchsia x86_64
• Illumos
• iOS aarch64
• iOS x86_64
• Linux aarch64
• Linux arm gnueabi
• Linux arm-musleabi
• Linux armv7 gnueabihf
• Linux i686
• Linux i686 musl
• Linux MIPS
• Linux MIPS64
• Linux MIPS64 el
• Linux mipsel
• Linux powerpc
• Linux powerpc64
• Linux powerpc64le
• Linux s390x
• Linux x32
• Linux x86_64
• Linux x86_64 musl
• Minver
• NetBSD x86_64
• OpenBSD x86_64
• OSX x86_64
• Redox x86_64
• Rust Stable

[SteveLauC]

Hi @asomers, looking at our source code, I am curious why this #[cfg_attr(docsrs, doc(cfg(all())))] is needed?

From my obervation:

1. We are not enabling the doc_auto_cfg feature, which means we won't get a #[doc(cfg(xx))] with #[cfg(xxx)], so no need to overwrite a non-existing value

2. The doc_cfg feature does not seem to work with assocaiated item:

   // src/lib.rs
   
   #![feature(doc_cfg)]
   
   #[cfg(target_os = "linux")]
   #[doc(cfg(target_os = "linux"))]
   pub struct Foo;
   
   impl Foo {
       #[cfg(target_os = "linux")]
       #[doc(cfg(target_os = "linux"))]
       pub const LINUX_SPECIFIC_CONST: usize = 1;
   }

   Run:

   $ cargo +nightly doc --target x86_64-unknown-linux-gnu --open

   That Available on Linux only. banner will be set for Foo only, not for LINUX_SPECIFIC_CONST

And some types in our code are partially annocated with this attribute, e.g., enum Request defined in src/sys/ptrace/linux.rs, which makes me confused

[asomers]

I think they're there because I originally developed the PR with the doc_auto_cfg feature (see 83ea2f3). When that's on, constants like O_NOCTTY get a "This is supported on non-Redox only" banner. Those banners are both annoyingly verbose and deceptive (see the linked commit message for the explanation for why they're deceptive). But then I must've removed doc_auto_cfg before I posted the PR, leaving the doc(cfg(all())) attributes useless. So we can remove them now.