Closes: #1739
Closes: #1740

* remove dump_age_period; not much added value compared to
  dump_age itself;
* remove dump_wm; dump_wm is a global constant that
  sets the maximal run size in a range; it must be a function of
  range size, which is defined per space; it's better to estimate
  it automatically (we may vary run size depending on the
  temperature of the run).
* remove dump_prio; we're abandoning zone-based scheduler,
  and this was merely a handle to prioritize dumps of runs in red
  zone;
* provide reasonable defaults for dump_age (40 seconds) and
  compact_wm (begin merge when there are at least two runs in a range)
* update tests

We take index->refs in these functions in order to ensure the index
won't be dropped while we're working with it in the tx thread. Actually,
this logic isn't specific to a particular backend and should be
implemented at the upper level. So let's drop it for now - this will
allow us to simplify vy_index reference counting, which we do need to
avoid use-after-free in workers.

sdc is a buffer allocated in each worker thread and passed to worker
functions where it can be used to avoid malloc. Turns out it is only
used in vy_range_redistribute_index(), where only sdc->b (vy_buf) is
needed. And even that is dubious, because it isn't a hot path, so it'd
be OK to call malloc there directly. So let's kill this sdc struct.

Closes #1736

Allow to disable logrotate. Closes #1746.

Needed for Gentoo ebuild.

systemd-tmpfiles will care about that.

Needed for Gentoo ebuild.

Caused by #1527

Make this code to be more clear.

Keep comments in one place. When we move the forward declarations
to a header file, the comments will migrate along.  For now, keep
them near function definitions: they'll get improvements, and it'd
be difficult to maintain two copies otherwise.

* Implement new merge_iterator
* Implement new txw iterator
* Implement read iterator

* remove range->reflock

dump_age_wm, set to 1MB, was a watermark triggering dump of any
run with size exceeding 1MB.

Technically, dump_age scheduler was involved only after compact/
dump scheduler had their turns. But since we're moving to a single
global priority queue for all ranges, keeping this watermark
would put age dump scheduler squarely at odds with gc scheduler.

This is perhaps the last chunk of renames eradicating Sophia
name legacy.

There is no way to delete from stailq.

Eliminate multi-thread access to vy_quota and replace pthread_cond with
fiber-friendly ipc_cond.

Fixes #1601

It turns out that env->seq->vlsn (view sequence number, needed for
taking a consistent snapshot) is updated exclusively in the tx manager
thread and used only in vy_task work functions, which are scheduled by
the manager itself. So there isn't much point in keeping this variable
in the global environment and protecting it with a mutex - instead we
can store it directly in tx_manager and simply pass it as a member of
vy_task struct.

This patch isn't intended to introduce any functional changes.

Closes #1687

Decimal number for lsn in index name.
Don't lock index on checkpoint commit.
Link index name when wait for checkpointi lsn.

Also use lbox_fiber_cancel both for cancel and kill since it's the
strict superset of both.

Closes Debian/#835798 (FTBFS).

* renames
* use realloc(), not malloc + copy loop
* don't allocate memory in constructor
* follow Tarantool naming conventions for constructor/destuctor
  (create/destroy)
* update tests
* reduce test case sizes
* reduce valgrind varnings in the test case

Follow up #1731