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.

Closes #1811

Convert requeset.cc to request.c (plain C) to make
request_normalize_ops() callable from C. Move all prepareatory
work from Vinyl_Space::executeUpsert() to vy_upsert() to make
vy_upsert() callable from other contexts.

Encapsulate vy_insert() in vy_replace(), to make vy_replace()
a single entry point to all kinds of replaces. This is necessary
to make vy_replace() easily usable in other contexts.

Remove VinylEngine::recovery_complete, which was fully
dependent on vy_env::status. Provide public access to
vy_env::status.

In preparation for folding of memtx_replace_build_next() in scope
of recovery refactoring, mvoe all replace functions to memtx_space.cc

No semantical changes.

Don't count snapshot rows when recovering bootstrap.snap.
Add a comment to memtx_add_primary_key()

In case of multi-level compaction, we compact and hence delete not all
runs, so we must not delete the runs left on the list from the metadata
log, but we do now. This breaks recovery. Fix this.

Add error injection for vinyl schedule timeout. Used for vinyl
error injection tests.

New test for vinyl checkpoint

Before dumping or compacting a range, consider coalescing it with its
neighbors. Only coalesce two adjacent ranges if the total size of disk
space and in-memory data occupied by them is <= range_size / 2, to avoid
split-coalesce oscillations.

Closes #1735

Rename tuple.cc, tuple_convert.cc, tuple_format.cc

No semantic changes.

* Re-order code in tuple.h, tuple_format.h and key_def.h
* Add _xc prefixes to tuple.h functions
* Fix an unchecked exception in box_tuple_iterator()

Closes #2038

Add helpers to decode numbers and use them throughout the code.

tuple_update/upsert was inlined in memtx_space and
box_tuple_update/upsert.

Fixes #1778

Needed to ensure that update_optimize will work
after implementation of common memory level
for txv and vy_mems.

Needed for #1908

Exactly the same piece of code (including comment) used by both dump and
compact. Factor it out into a separate function.

All the three functions that need a write iterator, dump, compact, and
split, reuse the same code for initialization of the write iterator.
Factor it out into a separate function.

* Use SNPRINT() macro from util.h
* Prettify formatting with missing schema (e.g. "user@localhost")
* Add Lua test case

Fixes #1924

1024MB is a lot for small devices and virtual machines.
256 MB is more than sufficient to play with a database,
and will make life easier when trying the database for
the first time.

Closes #1692

Change order of traversal in merge iterator to direct,
getting ready for avoiding reading from disk in case of cache hit