Use underscores instead of dashes in file names.
Use 0 for success -1 for error.

Update code and tests.

With snapshot recovery, we have 3 different states of
replication:
- no replication at all
- there is a replication source, but no connection to the
master
- connecte to master and reading snapshot/xlog

This is a preparatory step to (in future) separate
setting of replication source and establishing a connection
to the master.

Conflicts:
	extra/rpm.spec.in
	extra/rpm/tarantool.rpm.spec.in
	test/CMakeLists.txt

node -> server
lsnmap -> vector clock
setlsn -> vclock

Simplify replication relay code. Rename a global variable
to a name that doesn't match a type name (simplifies debugging).

There are 3 cases of vclock update:
- add a new server (vclock_add_server())
- get the next lsn for a local server (vclock_inc())
- update a vclock copy from some remote master (vclock_follow()).

Extract and document vclock merge logic.

Instead of manually adding a surrogage server id for snapshot,
add it to the cluster clock when saving a snapshot.

vclock.h is now a C header, and in clang C you can't
declare a structure in a for loop.

During bootstrap, work even if server_uuid/cluster_uuid are
present in the snapshot. This is necessary to be able to upgrade
bootstrap.snap over time.
As a proof of concept, rewrite bootstrap.snap from an output snap.

Begin counting lsns from 1, not from 2.

Fixes gh-352

It takes an incomprehensible piece of code which juggles with
master-master state (vector clock), and turns it into another
hardly comprehensible piece.

However, all tests pass, so this alone must justify the changes.
Another reason is that they are made by my blessed hand,
thus destined to be better even than sliced bread.

On a more serious note it's an attempt to more carefully
control vclock state changes through a set of select calls:

- add a new server to vclock
- "follow" local clock with an update from a remote vclock;
supposedly this update is dealing with an authenticated
server, so less checks are necessary, and an error in vlcock
order is an indicator of a programming error or corruption,
so should lead to panic, not to an exception
- a call to get the next LSN for a local coordinate of vclock.

This patch also renames vclock.cc to vclock.c to make it
FFI friendly.

Make an explicit call to vclock for adding
a new server.

Word police: use server rather than "node".

Move stat_cleanup(), which resets the statistics
skewed by recovery to the place after recovery_finalize(),
to take into account records from remaining WAL files
which are read in recovery_finalize().

Rewrite stat.test to not use space with a system
id 0.

The test fails sporadically when running as part of a suite.

Conflicts:
	extra/rpm/tarantool.rpm.spec.in

Conflicts:
	extra/rpm/tarantool.rpm.spec.in

Conflicts:
	extra/rpm/tarantool.rpm.spec.in

Conflicts:
	extra/rpm/tarantool.rpm.spec.in