Currently, on recovery we create and load a new vy_run per each slice,
so if there's more than one slice created for a run, we will have the
same run duplicated in memory. To avoid that, maintain the hash of all
runs loaded during recovery of the current index, and look up the run
there when a slice is created instead of creating a new run.

Note, we don't need to do anything like this on initial join, as we
delete the run right after sending it to the replica, so we can just
create a new run each time we make a slice.

In order to recover run slices, we need to store info about them in the
metadata log, so this patch introduces two new records:
 - VY_LOG_INSERT_SLICE: takes IDs of the slice, the range to insert the
   slice into, and the run the slice is for. Also, it takes the slice
   boundaries as after coalescing two ranges a slice inserted into the
   resulting range may be narrower than the range.
 - VY_LOG_DELETE_SLICE: takes ID of the slice to delete.

Also, it renames VY_LOG_INSERT_RUN and VY_LOG_DELETE_RUN to
VY_LOG_CREATE_RUN and VY_LOG_DROP_RUN.

Note, we don't need to keep deleted ranges (and slices) in the log until
the garbage collection wipes them away any more, because they are not
needed by deleted run records, which garbage collection targets at.

The same keys will be used to specify slice boundaries, so let's call
them in a neutral way. No functional changes.

Currently, there can't be more than one slice per run, but this will
change one the single memory level is introduced. Then we will have to
count the number of slices per each run so as not to unaccount the same
run more than once on each slice deletion. Unfortunately, we can't use
vy_run->refs to count the number of slices created per each run,
because, although vy_run->refs is only incremented per each slice
allocated for the run, this includes slices that were removed from
ranges and stay allocated only because of being pinned by open
iterators. So we add one more counter to vy_run, slice_count, and
introduce new helpers to be used for slice creation/destruction,
vy_run_make_slice() and vy_run_destroy_slice(), which inc/dec the
counter.

There's a sanity check in vy_range_needs_split() that assures the
resulting ranges are not going to be empty: it checks the split key
against the oldest run's min key. The check is not enough for the slice
concept, because even if the split key is > min key, it still can be <
the beginning of the slice.

We use run->info.keys to estimate the size of a new run's bloom filter.
We use run->info.size to trigger range split/coalescing. If a range
contains a slice that spans only a part of a run, we can't use run->info
stats, so this patch introduces the following slice stats: number of
keys (for the bloom filter) and the size on disk (for split/coalesce).
These two counters are not accurate, they are only estimates, because
calculating exact numbers would require disk reads. Instead we simply
take the corresponding run's stat and multiply it by

    slice page count / run page count

There will be more than one slice per run, i.e. the same run will be
used jointly by multiple ranges. To make sure that a run isn't accounted
twice, separate run accounting from range accounting.

Make sure that we start iteration within the given slice and end it as
soon as the current position leaves the slice boundaries. Note, the
overhead caused by extra comparisons is only incurred if the slice has
non-NULL boundaries, which is only the case if the run is shared among
ranges.

When we finally move to single memory tree per index (currently we
maintain one per range), dump will result in creation of a single run.
To add such runs to ranges and be able to iterate over statements that
are within a particular range, we introduce a new concept, a run slice.
A run slice is a simple object that references a run and contains begin
and end keys inherited from the range it belongs to. Runs are now not
referenced directly by ranges, instead we use slices as an intermediary.
For now the concept is oversimplified: there may only be one slice per
run, but the following patches will remove this limitation.

Maintaining range->{begin,end} as tuples is useful for the concept of
run slices, which is introduced by the following patches. A slice may
inherit its begin and end from a range, so basically we have two
alternatives: either copy keys or take references. The latter seems to
be more straightforward.

The vinyl/recover test was written long time ago. Back then recovery was
a convoluted procedure based on scanning the data directory so that it
was crucial to generate a number of stale files to check its validity.
Nowadays, there's no point in it thanks to the metadata log. Moreover,
when the single memory level is introduced, ERRINJ_VY_RANGE_SPLIT won't
make sense any more as range splitting will not involve a worker thread.
So we can remove errinj from this test.

Another problem with this test is that it doesn't take into account data
compression (all tuples generated by it are compressed perfectly).
Handle this by generating random padding strings.

Also, when we snapshot vinyl stats before restart, there still may be
compaction in progress that will modify the stats in the last moment,
resulting in a sporadic failure. Address that by checking stats
separately, for another space with compaction disabled.

This reverts commit f3764063.

With the recently proposed concept of run slices, this commit is not
needed to implement the single memory level any more.

Conflicts:
	src/box/vinyl.c
	src/box/vy_log.c
	src/box/vy_log.h

This reverts commit f3ecce75.

With the recently proposed concept of run slices, this commit is not
needed to implement the single memory level any more.

Conflicts:
	src/box/vinyl.c

This reverts commit 55685eaf.

With the recently proposed concept of run slices, this commit is not
needed to implement the single memory level any more.

Conflicts:
	src/box/vinyl.c

Patch fb9c8b32 introduced
generation of reproduce code and dump of it to the log. But
the problem is that the code is initially generated in a big
lua string using repeated concatenation in a loop. Such a use
of lua strings is too vulnerable in terms of performance.
Avoid repeated concatenation of lua string in tx_serial.test.

This reverts commit 44367f5c.

With the recently proposed concept of run slices, this commit is not
needed to implement the single memory level any more.

Conflicts:
	src/box/vinyl.c

This reverts commit a839af29.

With the recently proposed concept of run slices, this commit is not
needed to implement the single memory level any more.

This reverts commit 1360f193.

The idea behind this patch makes sense for memory iterator, but the
implementation is incorrect: it's not enough to patch ->restore, because
memory iterator can start right from ->next_key, bypassing ->restore.
Since other hunks of this patch are not needed in the scope of the run
slice paradigm, and what is done by this patch needs to be rewritten
from scratch anyway, revert the whole patch.

Fixes commit 976b31cb ("vinyl: fix order error in txv_iterator_start").

If search key is empty, txw iterator starts from the first or last entry
in the write set depending on the iterator direction. This is incorrect,
because the write set is grouped by index so if the first/last entry
happens to be for another index, txw iterator will stop immediately even
if there are statements for the given index in the write set. Instead we
must take the first/last statement for the given index. We can use
psearch/nsearch for it - it will position to the first/last element in
the tree equal to the search key; since the search key is equal to any
statement of the given index in case the given key is empty, it will do
the job.

While we are at it, also remove handling 'key == NULL' case from the
write_set_key_cmp() as it is not used anywhere.

Update snapshot timestamp if a snapshot already exists.
Do not produce error if the snapshot already exists.
Fixes gh-2045.

Use one function to initialize a global read view, pass read
view lsn in.

The upsert squash process must not touch or take into consideration
non-committed statemens because they are owned by TX manager and
thus might be changed or removed by it's will.
Add a committed read view in TX manager (in addition to global
read view) that allows to see only committed statements.
Use the read view in squash process.
Move upsert count calculation, simple inplace upsert
squash and upsert process invocation from preparation stage to
commit stage of TX; without this change after a squash process
prepared statements will not get a change to use simple inplace
upsert squash.

fix gh-2382.

Due to some historical reasons, vy_upsert hides some fatal errors
in the upsert statement and returns the original base statement
with original lower lsn. In that case a squash process replaced
wrong statement with wrong lsn doing usless work. Fix it.

There might be a case, when a non-upsert statement was inserted just
after a sqush process of the same key was started. Before this patch
the squash process copied that non-upsert statement and reinserted
it into the index for some reason.
Make the squash process to detect the case and quit immediately.

We are not ready to freeze `box.info().vinyl()` and `index:info()` output
right now. Let's keep this API for internal use only. Please note that
the output might change in the future.

See #1662

* Rename box.info.server.id to box.info.id
* Rename box.info.server.uuid to box.info.uuid
* Rename box.info.server.lsn to box.info.lsn
* Rename box.info.cluster.signature to box.info.signature
* Drop box.info.server section
* Return `nil` instead of `0` for box.info.id during bootstrap
* Return `nil` instead of `-1` for box.info.lsn during bootstrap

Sample output:

```
tarantool> box.info
---
- version: 1.7.3-538-g7ee75dee4
  id: 1
  ro: false
  vclock: {}
  uptime: 3
  lsn: 0
  vinyl: []
  pid: 20714
  status: running
  uuid: e6d913b9-a8b8-4873-ba94-14cf4357fec6
  signature: 0
  replication:
    1:
      id: 1
      uuid: e6d913b9-a8b8-4873-ba94-14cf4357fec6
      lsn: 0
  cluster:
    uuid: 3cfa8749-0fba-4b13-b8eb-db90f79d1485
```

Closes #723

Prepare for box.info.server removal.

See #723

page_index_offset depends on the XLOG header size, which now contains
PACKAGE_VERSION, which obviously is of variable length. Moreover,
the output of `git describe` used by PACKAGE_VERSION may have different
length on various systems even for the same commit.

See https://github.com/packpack/packpack/issues/72 for details.

Follow up ab5b6488

A fiber_sleep() triggered by ERRINJ_WAL_SHORT_DELAY in txn_commit()
caused invalid commit order for transactions. Rewrite the test case
to use ERRINJ_WAL_DELAY from wal_write_to_disk() instead.

Other changes:

 * Refactor box/iproto_stress.test.lua to use modified ERRINJ_WAL_DELAY.

Follow up a49f6711

* Get rid of the preprocessor magic
* Add ERRINJ_DOUBLE
* Remove unused code

Use ERRINJ_WAL_WRITE instead.

Follow up a49f6711

* Don't use string.format() in case of absence of vararg
* Add pcall() for string.format() and propogate error to upper levels
  for better debugging

Fixes #889

* Add `Version: 1.7.x-nnn-gXXXXXX` to XLOG header
  to use in the future versions.
* Add the test cases for XLOG header.

See #2100 and #2297

* delete background fibers
* error on multiple run of test

* renames
* correctly account statement count in an index
* add comments