*)added proper cleanup
*)minor fixes and checks

Conflicts:
	src/iproto_port.cc

*) fixed potentionally unsafe code:
  on memory fail there could be port->found > 0 and uninitialized port->svp

 on memory fail there could be port->reply.found > 0 and uninitialized port->svp,
 that could lead to random error after svp restoration.
*) possibly fixed OPENTAR-97

finalizers.test marked as long run in suite

 *) added stress test for snapshotting
 *) moved finalizers.test.py to long_run

test-run udpated

set test-run module in gh-3 branch for review

closes gh-856

Conflicts:
	include/fiber.h
	include/mutex.h
	src/admin.cc
	src/admin.rl
	src/box/box_lua.cc
	src/box/request.cc
	src/box/txn.cc
	src/coeio.cc
	src/coio.cc
	src/fiber.cc
	src/ipc.cc
	src/iproto.cc
	src/log_io.cc
	src/lua/bsdsocket.cc
	src/lua/init.cc
	src/lua/session.cc
	src/memcached-grammar.cc
	src/memcached-grammar.rl
	src/memcached.cc
	src/recovery.cc
	src/replica.cc
	src/replication.cc
	src/say.cc
	src/session.cc
	src/tarantool.cc
	src/util.cc

Rename global `fiber` into `fiber_ptr`

Rename global `fiber` into `fiber_ptr` (for use with gdb and avoidance of ambiguity with `struct fiber`)

1. Enable background=false for test-run
2. Check local directory for config

*) fixed bsize() method of bitset index
*) added macro switch to revert bitset index changes to previous state
*) the macro is not defined by default - new version of bitset index un use

Conflicts:
	src/module/mysql/mysql.cc
	test/module/net_sql.mysql.result
	test/module/net_sql.mysql.test

Backport fix from tarantool/mysql by Anton Reznikov

Implement box.once().

Right now there is a race in box.once(), this is somethign we'll
need to fix when we have master election on simultaneous bootstrap.

The patch merely "establishes" the syntax which is going to work
most of the time.

When handling join, an error from box_on_cluster_join()
was ignored and not sent to the client, so the client
believed that the JOIN was handled successfully and proceeded
with SUBSCRIBE.

Before sending OK to join, execute box_on_cluster_join(),
if it fails, the client sees an error and stops.

Make ER_REPLICA_MAX a logged error.

No test case since the test requires a fail-stop test of the replica,
which is clumsy in our framework.