* add TokuDB, together with the ft-index library
* cmake support, auto-detecting whether tokudb can be built
* fix packaging - tokudb-engine.rpm, deb
* remove PBXT
* add jemalloc
* the server is built with jemalloc by default even if TokuDB is not built
* documentation files in RPM are installed in the correct location
* support for optional deb packages (tokudb has specific build requirements)
* move plugins from mariadb-server deb to appropriate debs (server/test/libmariadbclient)
* correct mariadb-test.deb to be not architecture-independent
* fix out-of-tree builds to never modify in-tree files
* new handler::prepare_index_scan() method

It puts files in the correct location for rpm/deb packages.

After MDEV-4817, WHERE clause may be evaluated early, in fix_fields.
In ps-protocol fix_fields happens both for prepare and execute phase,
generating double the number of warnings from the early WHERE clause
executions.

* now only production-use server plugins are in mariadb-server.
  client plugins are now in libmariadbclient18,
  testing-only server plugins are now in mariadb-test
  (set Coflicts: field properly)
* fix Architecture: of mariadb-test
* terrible hack to support optional deb packages for optional plugins

pending merges:
  Alexander Barkov 2013-09-09 MDEV-4863 COALESCE(time_or_datetime) returns...

MDEV-4941 make: AIX fails with 'Identifier not allowed in cast'; syntax error in include/my_global.h

C++ comments in C files, and a typo in my_global.h

cmake/plugin.cmake:
  * auto-create an rpm for a plugin, if it places itself in a new component
storage/tokudb/CMakeLists.txt:
  install tokudb in COMPONENT tokudb-engine.
  this automatically creates a separate rpm for it.

* disable jemalloc on windows (cannot run ./configure)
* disable jemalloc on ancient cmake (ExternalProject does not work)
* rewrite TokuDB compiler test to check for features, not versions (to work on cmake before 2.8.11)
* fix ft-index to not add VALGRIND_INCLUDE_DIR to includes, if no valgrind was found
* correct the package name in FindValgrind.cmake (for find_package(... REQUIRED) to work)
* disable ft-index tests by default (faster compilation and they aren't used anyway)
* don't build ft-index with valgrind by default (otherwise it *requires* valgrind, it doesn't auto-detect)
* use --loose-tokudb in the .opt file

(otherwise one cannot even pull on windows)

mysql-test/CMakeLists.txt:
  don't generate collection fules. we don't use them
  and they break out-of-source builds

tell the optimizer that every TokuDB "clustering" index is the "covering" index
in the MariaDB optimizer sense.

While it is stable on itself, the MariaDB integration
within 5.5 series is rather new.

cmake/jemalloc.cmake:
  for dependencies to work, LIBJEMALLOC should be the target name, not the path
storage/tokudb/CMakeLists.txt:
  * check the preconditions
  * disable bdb tests (compilation errors)
  * set variable, instead of SET_PROPERTY. same effect,
    but doesn't fail when a plugin is disabled (that is, a target does not exist)
storage/tokudb/ft-index/CMakeLists.txt:
  cmake should not look into examples/ directory,
  there is hand-crafted examples/Makefile that
  cmake will overwrite
storage/tokudb/ft-index/buildheader/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/cmake_modules/TokuMergeLibs.cmake:
  Libraries must be specified in the specific order,
  REMOVE_DUPLICATES cannot be used, because it destroys this order.
  
  (when OSLIBS contains "-lpthread -ljemalloc -lpthread", REMOVE_DUPLICATES
  makes it "-lpthread -ljemalloc". But a thread library *must* be *after* jemalloc)
storage/tokudb/ft-index/cmake_modules/TokuSetupCTest.cmake:
  * 'which' might print errors to stderr, they are not important, shut them up
  * we don't have TOKUDB_DATA, no need to warn about it
  * don't configure_file into itself (with input=output)
storage/tokudb/ft-index/cmake_modules/TokuThirdParty.cmake:
  jemalloc is built externally to tokudb/ft-index
storage/tokudb/ft-index/ft/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/ft/tests/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/locktree/tests/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/portability/CMakeLists.txt:
  s/jemalloc/libjemalloc/
storage/tokudb/ft-index/portability/os_malloc.cc:
  unnecessary include file
storage/tokudb/ft-index/portability/tests/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/src/tests/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/util/tests/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name
storage/tokudb/ft-index/utils/CMakeLists.txt:
  the syntax is ADD_EXECUTABLE(target source) and "source" is the file name

     file_logger became the service.
     Data like query_id now are sent to the audit plugin.
     Fix for MDEV-4770 ported from 10.0.
     Fix added for the read_maria_plugin_info().
     Log rotation can be disabled with 'set rotations=0'.

                TOKUDB_ZLIB | TOKUDB_QUICKLZ |
                TOKUDB_LZMA | TOKUDB_FAST |
                TOKUDB_SMALL }
syntax for tables

The fix for bug mdev-4971 not always correctly set the pointers to
inherited multiple equalities in objects of the Item_equal class. 

            ORDER BY does not work

Use "dynamic" row format (instead of "block") for MARIA internal
temporary tables created for cursors.

With "block" row format MARIA may shuffle rows, with "dynamic" row
format records are inserted sequentially (there are no gaps in data
file while we fill temporary tables).

This is needed to preserve row order when scanning materialized cursors.

If rpl_semi_sync_master_timeout is large, calculation of absolute waiting time in semi-sync plugin is inefficient. This error is specific to systems with 64 bit long values (all 64 bit Unixes)
In  rpl_semi_sync_master_timeout has maximal value (= MAX_ULONGLONG), calculating abstime may require  ~ 18 billion subtract operations.

The fix is to use division instead of subtraction-in-a-loop. Also fixed an integer overflow bug.