Previously we used a pessimistic approach with a full motion for
"not in" operator. But after some research it seems that we can use
the same logic as we do for equality and "in" operator, i.e. choose
either local, either segment or either full motion policies.

Closes #226

It was a wrong assumption that `a <> 1 and a <> 2)` is equal to
`(a, a) <> (1, 2)`. Fixed.

As a side effect some refactoring was made to deal with code
duplication in the cartridge, mock and bench engines.

Previously, only the table cache was invalidated on the router. Now
we flush cached IRs as well.

Thanks to the updates in tarantool module we don't use tarantool
symbols to work with decimal. As a bonus we can remove our mocking
framework with dynamic linking of the decNumber library to cargo
test binary.

As a side effect the code of explain was refactored and simplified.

Otherwise we face an error during c-ares submodule update in make
init target.

Previously, we created a static archives of the msgpuck and decNumber
libraries and made a statically linked them into the test executable.
After tarrantool module migrated to dlsym, we can no longer use static
linking.

As a result we build shared libraries for msgpuck and decNumber to
dynamic link them into the unit test binary.

As a side effect we got rid of the environment variable in a unit
tests build. Now we rely of a feature "mock" for this purpose.

1. When we build a configuration cache and face a table without
   a sharding column, simply skip this table (do not exit with
   error).
2. Make "space not found" error more debug firendly
3. Rename "broken test" to "ddl"

BREAKING CHANGE: reqiures Picodata Tarantool with a string quoting
fix (e9360c989129a755b86486c7bcd6e2d21a078e22).

We should not expose a sharding column (i.e. "bucket_id") for any
parts of the join if the user doesn't mention it explicitly in
projection. Otherwise we receive incorrect query results.

The error was produced by the assumption that (2, 1) > (1, 2) is
equivalent to (2 > 1) and (1 > 2). That assumption was incorrect,
so we don't merge AND chains for any operators other than Eq and
NotEq.

BREAKING CHANGE: dispatch_query() wraps result with a tuple of a
single element.

We made a wrong enumeration for the new columns in projection
when a system sharding column ("bucket_id") was located somewhere
except the last position in the relation schema. Fixed.