Currently, a dropped index might have compaction scheduled on scheduler
stop, in which case its ranges are inter-linked and cannot be easily
freed (dumb free of all ranges would result in double free). So let's
drop the final index cleanup for now. We need to rework it anyway,
because we never free active indexes on shutdown, although we should to
make valgrind happy.

Currently, compaction works as follows:
 1. worker: write old range's runs and shadow indexes to disk, creating
    as many new ranges as necessary (min 1)
 2. tx: redistribute active memory index of the old range among new
    ranges
 3. tx: replace old range with new ones and delete old one

Such a design has a serious drawback: redistribution (step #2) scales
lineary with the number of tuples and hence may take too long.

So this patch reworks the compact procedure. Now it looks like this:
 1. tx: create new ranges and insert them to the tree instead of the old
    range; in order not to break lookups, link mem and run of the old
    range to new ones
 2. worker: write mem and run of the old range to disk creating runs for
    each of new ranges
 3. tx: unlink old range's mem and run from new ranges and delete it

An old range is either split in two parts by the median key
(approximate) or not split at all, depending on its size.

Note, we don't split a range if it hasn't been compacted at least once.
This breaks assumptions of vinyl/split test, so I disable it for now. I
will rework it later, perhaps after sanitizing the scheduler.

Closes #1745

Reorder the branches of squash loop to begin with REPLACE,
this makes the reasoning about this loop a whole lot easier.

Always use vy_tuple_key_part(key, 0) == NULL
as an indicator of NULL key.

Before this patch, some of the edge case optimizations in
run, mem and txv iterators didn't work, since they compared
input key with NULL, whereas it was set to a special key with zero
parts.

@todo: this is a yet another discrepancy stemming from distinct
tuple formats in memtx and vinyl.

Index page size and range size are storage-level properties
and can not change during index life time.
Initialize these properties from global defaults at index
creation time. This way they are not affected after restart
with different global options.

Remove "engine-level" options support from Lua, all options
are known to us now that we ditched third party engines.

Closes #1738

Closes #1737

g/static inline/statis is easier than merging an old branch
with a runaway trunk

ops

Do a proper roll back on dump/compact failure.

Now, names of related interfaces, global variables and
fields are based on "iterator".

This is pure refactoring without any functional changes.

Closes #1694

The range iterator open function used an obsolete indicator
for special key (NULL key), which led to selection of a wrong
range in :max() query.

The bug required a compaction process to trigger it, since
when there was only 1 range max() naturally always got the last one.
After compaction, range iterator open could yield a random
range for the special key.

Update errinj.test.lua to always trigger compaction.

Remove all ranges from scheduler before deleting index.

Follow up 43ca199e

The merge iterator didn't take into account iterator diretion
when merging its sources, and always used to return the smallest
tuple among the alternatives.

If iterator order is LT or LE, the biggest tuple should be returned
instead.

We don't remove dead index's ranges from the scheduler heap after we
scheduled a drop task for it - we only do it from the drop task itself.
As a result, we can happily go and schedule another task for the index
being already scheduled for destruction. This can result in the
index->ref == 1 assertion being violated in vy_task_drop_execute() or
memory corruption induced by use-after-free of a vy_index struct.

Switch to pointers from built-in structures range->i[01].

* add a test case based on error injection
* fix a few release mode warnings
* the test case is still failing, depends on introduction of shadow
memory indexes.

Use global queues for scheduling instead of per-index.

Fixes #1708

Update test run (fixes the sporadic console handshake bug when running
tests).

Fixes gh-1163.

Closes #1599