tarantool_box without arguments runs in interactive mode.
@todo:
    - log_level by default should be 'critical'
    - wal_mode by default should be 'none'
    - pid_file by default should be 'none'
    - readline should be called via coeio, to not block
    event loop while it's running

See rationale for this fix in gh-223
Remove on the same token a buch of dead code for sandboxing
and do a few renames.

In preparation of removal of confetti and switching solely to
Lua-based configuration file, reduce dependencies on struct tarantool_cfg
and reorder server initialization, so that it can happen after the event
loop has already been started.

Remove command line options:

--cfg-get
--check-config
--config
--init-storage

- these are gone forever, in preparation for config-less start.

--background --  move to tarantool.cfg for now, in the future
it will be part of Lua config.

Make #! mandatory - if the server is started without #!, it's
only possible to query its version or get startup help (--version
and --help are the only remaining supported options).

When #! is present, do not parse options, pass them wholly to the
script at hand.

Add helper scripts to test suites, to start/stop the server.
For now, the scripts are empty, in the future they will be
merged with a respective config file.

Conflicts:
	doc/user/lua-tutorial.xml
	doc/user/plugins.xml
	doc/user/stored-procedures.xml

 * Change format of xlog/snap files to MsgPack-based format
 * Update replication to use IPROTO protocol
 * Rework replica bootstrap to process snapshot by rows
 * Add support for composite rows in fio_batch
 * Fix iproto->sync in SUBSCRIBE command

Conflicts:
	doc/user/data-model.xml
	doc/user/tutorial.xml

Conflicts:
	debian/tarantool-sql-module.install
	extra/rpm.spec.in
	src/lua/init.cc
	test/CMakeLists.txt

refactory.

* Rewrite box_net.{pg,mysql}.test.lua, so line wrappings won't matter.
* Testing System: 'RelWithDebugInfo' closer to Release, than to Debug
* Compiling on clang -> additional OP, cause of garbage after
  `iproto_request_type_strs` (Failed admin and other tests)

When panic() is called from a fiber != sched, we
call atexit handlers and free all memory, including
runtime memory, including the memory used for fiber
stack.

Thus essentially freeing memory the fiber's stack is
using.

The very next instruction using the stack gets
a SIGSEGV afterwards.

Perhaps don't be so strict about freeing resources at
shutdown and do not shoot yourself in the foot.

Conflicts:
	test/module/net_sql.common.lua
	test/module/net_sql.common.result
	test/module/net_sql.common.test
	test/module/net_sql.mysql.lua
	test/module/net_sql.mysql.result
	test/module/net_sql.mysql.skipcond
	test/module/net_sql.mysql.test
	test/module/net_sql.pg.lua
	test/module/net_sql.pg.result
	test/module/net_sql.pg.skipcond
	test/module/net_sql.pg.test
	test/test-run.py

purpose: OOS-build tests fail

Conflicts:
	.travis.yml
	client/tarantool/tc.c
	client/tarantool/tc_pager.c
	src/lua/yaml.cc

Needed for server/pg_type.h header

CMake is shipped with broken PosgreSQL script in most distros.
I had to copy this script and patch it to properly support
new versions of PostgreSQL.

 * Install *.lua modules into ${DATADIR}/tarantool/1.5
    and *.so modules into ${LIBDIR}/tarantool/1.5
 * Fix default package.path and package.cpath
 * Install Lua headers to ${INCLUDE_DIR}/tarantool/1.5
 * Install config.h to ${INCLUDE_DIR}/tarantool/1.5
 * Add a new package - tarantool-dev (server development headers)
 * Use RPM Macroses in rpm.spec
 * Fix tarball generation for Debian
 * Use GNUInstallDirs.cmake to setup directories

purpose: OOS-build tests fail

to support multiple engines.