Jenkinsfile is a script, akin to .travis.yml that contains
instructions how to build code in the repo. I'm experimenting
to see if we can implement a similar workflow to what we have
on travis.ci

Totally disable /usr/lib/rpm/brp-strip-* scripts on all
versions of rpmbuild.

Fixes #1901

Caused by a typo: range -> result

Closes #2057

To improve incapsulation, make vy_recovery struct private to vy_log.c
and provide vy_log methods needed for recovery.

* fix a bug with partial initialization of tuple_format
  in case of zero field count (space with no indexes, we still may
  allocate a tuple for this space, so need an initialized format,
  but will fail to insert it)
* replace assert() for small field_map_size with an error
  message, introduce a new error code for field count limit,
  i.e. limit on how many fields can participate in indexes
* renames & comments

Optimize tuple format for cases when all keys are sequential and
start from the first field:

    box.schema.space.create('memtx')
    box.space.memtx:create_index('sequential0', { parts =
        { 1, 'unsigned' }})
    box.space.memtx:create_index('sequential1', { parts =
        { 1, 'unsigned', 2, 'unsigned' }})
    box.space.memtx:create_index('sequential2', { parts =
        { 1, 'unsigned', 2, 'unsigned', 3, 'unsigned', 4, 'unsigned' }})
    box.space.memtx:create_index('random1', { parts =
        { 3, 'unsigned' } })

In the example above only field 3 is now indexed by offset (was 2, 3, 4).

+ Improve error message in field_type_create()

Closes #2046

 - VY_LOG_TX_BUF_SIZE: macro -> enum
 - Fix say_debug() format in vy_log_tx_commit()
 - Cleanup vy_log_flush_f() ret code convention
 - Add a separate function, vy_log_tx_try_commit(), instead of
   passing no_discard=true to vy_log_tx_commit()
 - Refactor 'if' statement in vy_index_drop()
 - Fix comments:
   * vy_log_flush(): why entire buffer is flushed
   * vy_log_open(): mention that we abort recovery if we can't
     flush the log
   * vy_log_tx_commit(): explain why the log can be not open
   * vy_log_tx_commit(): add comment stating that we rollback
     transaction on failure

Secondary index now store tuples with offsets.

Needed for #1908

Needed for #1908

* Remove unused part_count
* Unify arguments order

Needed for #1908

 - Name vinyl metadata log as <lsn>.vymeta, where <lsn> is the last
   snapshot signature.
 - Create a new log on checkpoint, discarding records left from
   dead indexes and records that cancel each other.
 - Log rotation is done asynchronously from a coeio thread so
   as not to stall ongoing transactions.
 - While the log is being rotated, all log writers are locked out.
   This should be OK, because log is rotated fairly quickly and
   the scheduler, which is the primary user of the log, runs rarely.

Closes #2031

We need to log index drop requests in order to be able to filter out
dropped indexes from the metadata log on rotation, which is yet to be
implemented. The problem is that we can't fail vy_index_drop(), because
index drop request has already been written to WAL by the time it is
called. We cope with it by writing VY_LOG_DROP_INDEX record with
@no_discard flag set to true, so that even if vy_log_tx_commit() fails,
it stays in the buffer and can be flushed along with the next
transaction. If it isn't flushed before the server is restarted, we
replay the record in the metadata log buffer on WAL recovery, returning
to exactly the same condition we had before restart.

While we are at it, rename VY_LOG_NEW_INDEX to VY_LOG_CREATE_INDEX for
consistency.

 - vy_log_write() now simply appends a record to the internal log
   buffer, hence it cannot fail, which simplifies the log API.
 - vy_log_tx_commit() flushes the buffer to xlog. It has a new flag
   @no_discard. If it is set, the buffer won't be reset on failure, so
   that the next transaction will retry to flush the records. Currently,
   it is not used by any caller, but it is required for a drop index
   record, which will be introduced later.
 - vy_log_write() can be used before the log was opened, in which case
   records will be flushed when the log is first opened. This is needed
   for drop records too.
 - vy_log_tx_rollback() is removed as it is not needed any more with
   vy_log_write() never failing.

In order to implement index drop logging, we will need to be able to use
vy_log during local recovery when its xlog hasn't been open yet (it is
open when recovery is complete). So let's extract the code opening xlog
file from vy_log_new() and put it in a separate function, vy_log_open(),
to be called after local recovery has completed, while creating vy_log
from vy_env constructor.

So that we can access next_run_id and next_range_id directly.
In order not to introduce dependency on latch.h, allocate a latch
from heap and use forward declaration.

Currently, vy_recovery_load_index() only calls the callback for
VY_LOG_INSERT_RANGE and VY_LOG_INSERT_RUN types of records, but not
for VY_LOG_NEW_INDEX, because the latter is not needed during index
recovery. However, we will need this event for vinyl metadata log
rotation, so let's make vy_recovery_load_index() call the callback
for this type of record too.

Not necessary as log records are checksummed.

net.box: set default timeout to 10s
tarantoolctl: set console connection timeout to inifinity

Closes #2054

closes gh-1925

* Don't start tarantool on install
* Don't restart tarantool on upgrade

Fix #2053

We don't pass the WAL error to the caller, so at least log it in the error
log.

Do not try to sync data range in case of write error in xlog_write_row().

Move fiber_pool struct to separate file. Prepare for detach fiber_pool
from cbus.

In order to rotate vinyl metadata log, we need to destroy the old xlog
and create a new one instead, so we can't embed the xlog structure in
vy_log. Let's store a pointer to xlog in vy_log instead.

* Remove "raw" comparators and dead code
* Fix inlining of comparison function
* Switch vinyl to specialized code from tuple_compare.cc
* Improve comments and minor renames

Prepare for #2046 and #2048

Remove struct vy_stmt * -> char * -> struct vy_stmt * conversion
from vy_task_compact_new().

We use ev_time() everywhere except xlog_init(), the latter uses ev_now()
for some reason. This prevents xlog from being created/opened from a
coeio task. Fix this.

* move cbus declaration to cbus.c, since going forward
  it's a private singleton
* rename cbus_item to cbus_endpoint. It represents the state
  of a single cord in the cbus (but in the future, a single cord
  may have more than one endpoint.
* move declaration of cbus_endpoint to cbus.c
* factor out cbus_find_endpoint()
* add/update comments
* move cbus_init()/cbus_free() to main.cc
* replace pthread_cond_signal() with pthread_cond_broadcast(),
  since in theory there can be many waiters on an endpoint
  to join.

Now we can create one-direction cbus connections and each cbus consumer
can have more than one producers. cbus_join will connect cord to cbus
with given name and cpipe_create will create a route to corresponding
destination

Fixes #2036

This is pure refactoring. No functional changes intended.