This change fixes two issues.  First, the current cfmakeraw()
implementation in this crate appears to be making a stack copy of the
input "struct termios" before modifying it, rather than correctly
modifying the original through the pointer.  Before this modification
the routine did not, thus, set the flags for raw mode.

Second, we address the default settings of the MIN and TIME terminal
options.  On at least FreeBSD and Linux systems, the modern default
value for MIN appears to be 1; i.e., block and wait for at least one
input byte.  On most Solaris and illumos systems, the MIN control
character slot overlaps with EOF, and thus has a default value of 4.
This breaks at least the examples in the "termion" crate, and probably
quite a lot of other software written first and foremost for Linux
systems.  We need to force the MIN value to 1 while switching to raw
mode.

References:
* NetBSD (became available with 8.0):
  http://cvsweb.netbsd.org/bsdweb.cgi/src/sys/sys/socket.h?annotate=1.129&only_with_tag=MAIN
  http://cvsweb.netbsd.org/bsdweb.cgi/src/lib/libc/sys/accept4.c?annotate=1.2&only_with_tag=MAIN
* Illumos:
  https://illumos.org/man/3socket/accept
  https://github.com/illumos/illumos-gate/blob/master/usr/src/uts/common/sys/socket.h
  https://github.com/illumos/illumos-gate/blob/master/usr/src/lib/libsocket/socket/weaks.c
* Solaris:
  https://docs.oracle.com/cd/E88353_01/html/E37843/accept-3c.html

Add AT_EACCESS to Linux, Solaris, and Fuchsia.

Add AT_SYMLINK_FOLLOW, AT_REMOVEDIR, and  _AT_TRIGGER to Solaris.

This triggers a warning on a recent nightly, which in turn breaks CI due to
`#![deny(warnings)]` in libc-test/build.rs

This is needed because Firefox now uses slice-deque rust crate.

It's *msg_prio* in both manpages and posix.

This constant is not stable across OS versions, so it cannot be used in
any backwards- or forwards- compatible way.  It's typically used to size
arrays in the kernel and in debugging utilities that are closely tied to
the OS version.  Since libc is ignorant about OS versions, we shouldn't
even be defining it.

These constants have already been deprecated for a few releases with
a deprecation notice, so they can finally be removed.

Closes rust-lang/libc#665

sigevent structs on most platforms have padding or unused fields. Rather
than display those in the Debug impl by deriving it, manually implement
all extra_traits instead ignoring those fields.

The second argument of `gettimeofday` was a `*mut c_void` on all targets,
but that type is incorrect in the following targets, where it should be
a `*mut timezone` instead:

On these other targets it appears that the signature of gettimeofday was incorrect (it takes a time-zone pointer instead of a void pointer):

linux+gnu: http://man7.org/linux/man-pages/man2/gettimeofday.2.html
freebsd: https://www.freebsd.org/cgi/man.cgi?query=gettimeofday&apropos=0&sektion=2&manpath=FreeBSD+11.2-stable&arch=default&format=html
openbsd: https://man.openbsd.org/gettimeofday.2
android: https://github.com/ricardoquesada/android-ndk/blob/master/usr/include/sys/time.h
dragonfly: https://www.dragonflybsd.org/cgi/web-man?command=gettimeofday&section=2

This commit corrects the type on these targets, which is a breaking change. Due
to how this API is commonly used (e.g. passing `ptr::null_mut` to the second
argument), breakage should be minimal. Users wanting to support both versions
can just write `ptr as *mut _` instead.

Closes #1338.