* Remove vclock_add_server()/vclock_has() - servers are implicitly
  registered in vclock->map on the first request
* As result, lsn == 0 is never exist in vclock
* box.info.vclock[server_id] returns nil instead of 0 even if
  server_id was registered in _cluster, but hasn't sent any rows
* Soften ER_UNKNOWN_SERVER checks for unregistered server_id
* Fix buggy ER_MISSING_SNAPSHOT logic
* Add missing "error.h" includes

Bump memory limit to 2GB

Allow to use the same configuration file on all cluster nodes for
replication.

Fix server_id vs server_uuid mess.

* Switch to YAML instead of JSON
* Add utilites for replication
* Print test-run internal errors to console

Store index as space_id:index_id. Configuration refactoring.

Move write operations out of index into separate
sophia_space.cc file.

Fix memory corruption caused by accidentally copy-pasted line.

space_truncate() depends on box_delete()/box_insert(),
move to appropriate place.

 - Introduce new two-phase JOIN protocol.
 - Remove Sophia snapshots and switch to checkpoints.
 - Include latest WALs and registration of replica into JOIN stream.
 - Remove legacy tests for read-only replica.

See box_proccess_join() for protocol changes.

Fill IPROTO_KEY for DELETE and IPROTO_TUPLE for INSERT/REPLACE.

Splitted init_field_map into init_field_map and tuple_validate