Add waitid and related constants and types.

[`waitid`](http://pubs.opengroup.org/onlinepubs/9699919799/functions/waitid.html) is a variation on `waitpid` with a marginally more convenient way of reporting the status, and a couple of handy
additional features, such as the ability to peek at an exit status without consuming it.  It's in POSIX.1-2008 and should be available on all supported Unixes.

Along with it come the type `idtype_t` and the constants `WEXITED`, `WSTOPPED`, `WCONTINUED`, and `WNOWAIT`.  The constants were already defined for unix/notbsd platforms.

The patch is currently incomplete: I'm pushing it to get CI to test platforms I don't have.  Todo list is

* [x] Add a definition of `siginfo_t` to all platforms that don't have it.
* [x] Verify that the new constants are consistent for all \*BSD platforms.
* [x] Verify that `idtype_t` is consistent across the board.
* [x] Add `link_name` annotations for `waitid` if/as necessary.

Add MSG_PEEK socket flag

#### MSG_PEEK
This PR exposes a cross-platform `MSG_PEEK` flag that a user may pass into the `flags` parameter for `recv()`, `recvfrom()`, or `recvmsg()` calls.

```
   MSG_PEEK
          This flag causes the receive operation to return data from the
          beginning of the receive queue without removing that data from the
          queue.  Thus, a subsequent receive call will return the same data.
```

In short, users may call `recv()` to "peek" at new data (or a message) without consuming it. Recall that in the normal case, calls to `recv()` consume the data from the socket's receive queue, making it unavailable to future calls.

#### Motivation
This change enables `libc` users, such as the rust standard library, to potentially add peeking functionality to TCP and UDP implementations, like `TcpStream` and `UdpSocket`, without being concerned about the highly platform-dependent nature of the flags.

(In this case, the flag's value `0x2` is very consistent, but that is not the case for many of the `MSG_*` flags. It makes sense to keep these differences confined to `libc`)

#### Reference
[bsd/apple: socket.h](https://opensource.apple.com/source/xnu/xnu-2050.7.9/bsd/sys/socket.h)
[bsd/freebsdlike/freebsd: socket.h](https://github.com/freebsd/freebsd/blob/master/sys/sys/socket.h#L418)
[bsd/freebsdlike/dragonfly: socket.h](https://github.com/DragonFlyBSD/DragonFlyBSD/blob/1f249c981c4e89e7cde1836a75b61cac36dc7ac5/sys/sys/socket.h#L367)
[bsd/netbsdlike: socket.h](https://github.com/IIJ-NetBSD/netbsd-src/blob/af5d253140491f2d1816c59ecb8a4d8a8d927688/sys/sys/socket.h#L517)
[unix/haiku: socket.h](https://github.com/haiku/haiku/blob/b65adbdfbc322bb7d86d74049389c688e9962f15/headers/posix/sys/socket.h#L114)
[unix/notbsd/linux: socket.h](https://github.com/torvalds/linux/blob/5924bbecd0267d87c24110cbe2041b5075173a25/include/linux/socket.h#L264)
[unix/notbsd/android: socket.h](https://android.googlesource.com/platform/development/+/73a5a3baaa5089f1ab2049e5934fa5d8a3f3e76a/ndk/platforms/android-20/include/sys/socket.h#229)

It turns out that *only* FreeBSD and NetBSD proper have waitid, and that
Solaris' additional W* constants are totally different from everyone
else's, which tips me over from 'find some way to shoehorn this into
the top-level unix/mod.rs' to 'put it in the submodules, live with the
code duplication'.

Remove kind='static' from Redox linkage

Our cross compiler links binaries statically, and the rustc target has dynamic linking disabled - adding kind = "static" is not necessary.

 * OpenBSD doesn't have idtype_t or the P_* constants either
 * FreeBSD has different values for the P_* constants
 * Android gives idtype_t a different signedness
 * Disable waitid on NetBSD as it causes a link failure - I think this
   may be a problem with the test environment

Add getpeereid function

This is a wrapper around getsockopt() for getting the uid/gid of a remote Unix domain socket peer. It was added in FreeBSD 4.6 and present in all modern BSDs I checked (including Mac OS X).

This is a wrapper around getsockopt() for getting the uid/gid of a
remote Unix domain socket peer. It was added in FreeBSD 4.6 and present
in all modern BSDs I checked (including Mac OS X).

 * idtype_t no longer an enum.
 * Darwin/x86-32 needs the $UNIX2003 thing.
 * Darwin, FreeBSD, and NetBSD all have different values for the new constants.
 * OpenBSD doesn't have this feature at all.  (Hopefully we can get away
   with defining idtype_t anyway.)

waitid() is a variation on waitpid() with a marginally more
convenient way of reporting the status, and a couple of handy
additional features, such as the ability to peek at an exit
status without consuming it.  It's in POSIX.1-2008 and should
be available on all supported Unixes.

Along with it come the type 'idtype_t' and the constants
WEXITED, WSTOPPED, WCONTINUED, and WNOWAIT.  Theconstants
were alre dy defined for unix/notbsd platforms.

Patch incomplete: several targets are going to have to add
definitions of siginfo_t, but I'm not sure which ones yet.

Link libc and libm in Redox [critical]

This is critical to being able to set up a proper cross compilation setup for Redox. Due to the way libc is vendored, I have to have this change merged in and have liblibc updated in the Rust repository as well before I can have a working buildbot for Redox.

Bump to 0.2.19

Add Solaris constants for fcntl-style advisory locking

Solaris doesn't implement flock(), so any Rust implementation of flock()
will need to implement it using fcntl(), using the F_RDLCK, F_WRLCK, and
F_UNLCK constants.

Solaris doesn't implement flock(), so any Rust implementation of flock()
will need to implement it using fcntl(), using the F_RDLCK, F_WRLCK, and
F_UNLCK constants.

Support Neg and Not in no_core mode.

Needed by rust-lang/rust#38776 which requires the traits to be implemented even for integer types.
This is already the case with binary operator traits, which always require the trait and its impls.

sparc64-linux support

This needs to be "cleaned" up to use modules instead of a bunch of `cfg`s ...

Sadly, sparc64 constants are very different from other architectures so cleaning this will result in a bunch of duplication, I think.

While working on this, I was wondering why the constants are not written like this:

``` rust
// linux/mod.rs
const COMMON: ::c_int = 3;

cfg_if! {
    if #[cfg(target_arch = "sparc64")] {
        const FOO: ::c_int = 1;
    } else if #[cfg(any(target_arch = "mips64", target_arch = "x86_64"))] {
        const FOO: ::c_int = 2;
    } else {
        // unsupported/unknown architecture
    }
}
```

I think this might result in less duplicated code. @alexcrichton Has something like that ^ been attempted before?

Added initgroups

Not sure whether this is the correct platform level. I'll wait for CI to find out..

EDIT: Okay now it builds on all platforms but I think this configuration does not add the function on BSD where it has the same signature as on Linux.

Add execl, execle, execlp

I tried to be as close as possible to the existing definitions of `execv` and friends.

btw: do you still welcome these PR? Doesn't the [new bindgen/build.rs method](http://fitzgeraldnick.com/2016/12/14/using-libbindgen-in-build-rs.html) offer to automate this crate to 100%

Added constants from prctl.h

I left in the comment from prctl.h I can remove them if you want.

Add binding for mremap.

Please consider this PR.

Let me know if  you need things changed.

Add memalign to Redox (for use in allocator)

This is required to satisfy a review item here: https://github.com/rust-lang/rust/pull/38401