Check evpipe[1] is the right way. Issue #1094.

* Check return codes of libyaml
* Synchronyze check_utf8() logic with libyaml
* Remove analyze_string() - this logic is performed by libyaml
* Add the test cases

Other changes:

 * Move CJSON sources to libserver to resolve dependency on lua/utils.c

gh-1249

move tarantool/sophia to tarantool-1.6 branch

* Split libyaml (C library) and lua-yaml (Lua bindings)
* Add support for system libyaml-dev in CMake scripts if
   -DENABLE_BUNDLED_LIBYAML is OFF (default is ON)
* Add libyaml-dev to DEB dependencies

third_party/libyaml/ is completely identical to upstream libyaml 0.1.6.
(libyaml-0-2 in Debian/Ubuntu). Test are passed both with bundled and
system versions of the library.

third_party/lua-yaml/ is patched by me to use luaL_serializer, therefore
upstream version is not suitable for Tarantool.

This patch is needed to pass Debian package checks (lintian).

Doing box.cfg{background=true, logger='/EACCESS.log'} resulted in a
silent failure because initialization routine closed stderr first before
initializing the logger.

Changing stuff to absolutize pathes (logger, pid_file) to make it easier
to troubleshoot errors.

Things in say.c (logger guts) were changed to call exit instead of _exit
on failure in order for the cleanup to be performed.

Also adding a library for managing pid files under
third_party/libutil_freebsd.

proctitle.c: since it's now fully rewritten by us, move from third_party
process_title.c: rename to title.c

Style changes.

 * add proctitle.h;
 * add config-time detection of setproctitle() syscall;
 * fixing program_invocation_(short)?_name globals (glibc);
 * fixing progname (getprogname(), initialized with argv[0]);
 * this module now works on osx;
 * fix osx ps displaying garbage (argc-1 extra \0 necessary);
 * general refactoring;
 * fix linux proctitle truncation (init_set_proctitle was called with
   argv+1, this is wrong);

 * CAVEAT: 'progname: ' prefix no longer present, will fix later

TODO: pass exact lsn as row.lsn to a slave during JOIN.
Ensure that slave properly accepts rows with variadic lsn.

Thanks Vyacheslav Egorov for the fix in LuaJIT.

Update third_party/luajit to v2.1 as suggested by Vyacheslav Egorov.

Wrong fix.

This reverts commit cd200c65.

Enable monotonic timers (if available).

Enable floor() usage in computations (instead of a crude approximation
with bithacks).

Enable eventfd and signalfd usage on Linux.

_Noreturn kw must appear before the function name as per C11 grammar and
clang7 is the first one to enforce that.

Ecb_noreturn macro is defined using _Noreturn keyword if C11 support is
detected; otherwise __noreturn__ gcc attribute is used. The later must
appear after the function parameters list.  These two methods to
annotate a function are inherently incompatible in terms of the token
location relative to the function declaration.

Use gcc attribute exclusively since it is always available.

_Noreturn kw must appear before the function name as per C11 grammar and
clang7 is the first one to enforce that.

Ecb_noreturn macro is defined using _Noreturn keyword if C11 support is
detected; otherwise __noreturn__ gcc attribute is used. The later must
appear after the function parameters list.  These two methods to
annotate a function are inherently incompatible in terms of the token
location relative to the function declaration.

Use gcc attribute exclusively since it is always available.

We already install signal handlers, but don't give them a chance
to execute, since don't yield during recovery. Add yields once
per a few tens of thousands of events.

Deploy a special hack with an idle watcher to make sure that
fiber_sleep(0) doesn't sleep for too long (1 ms without this
watcher).

libev sleeps for at least 1 ms (backend_mintime) unless there are
idle watchers. The logic of this behaviour is obscure to say the
least.

Since the upstream is also using *screw-the-community* cvs for
revision control, maintaining an own patch set on top of the
upstream is also a pain in the neck.

So, instead of patching libev (I hesitate to simply remove the
obscure logic since I don't understand the reason for it from the
comments, and I don't think making this logic optional is worth
the hassle, given also that we will diverge even more from the
upstream), create an idle watcher in fiber_sleep(), as a
workaround, to make sure backend_mintime is not looked at when
calculating sleep time for poll(), and zero timeout is used
instead.