ded545be9f7a416d1d50d653ab278dd0fc7e6c5e

Moved opt_for_system_time_clause from table identifier (which binds
it to single table) to table expression.

* BEGIN_TS(), COMMIT_TS() SQL functions;
* VTQ instead of packed stores secs + usecs like my_timestamp_to_binary() does;
* versioned SELECT to IB is translated with COMMIT_TS();
* SQL fixes:
  - FOR_SYSTEM_TIME_UNSPECIFIED condition compares to TIMESTAMP_MAX_VALUE;
  - segfault fix #36: multiple execute of prepared stmt;
  - different tables to same stored procedure fix (#39)
* Fixes of previous parts: ON DUPLICATE KEY, other misc fixes.

These features are defined only in XtraDB:
  chaged_page_bitmaps,
  innodb_changed_pages

* versioned DML: INSERT, UPDATE, DELETE;
* general refactoring and fixes.

Warning: breaks 'insert' and 'update' tests since they require part IV.

* moved vers_notify_vtq() to commit phase;
* low_level insert (load test passed);
* rest of SYS_VTQ columns filled: COMMIT_TS, CONCURR_TRX;
* savepoints support;
* I_S.INNODB_SYS_VTQ adjustments:
  - limit to I_S_SYS_VTQ_LIMIT(10000) of most recent records;
  - CONCURR_TRX limit to I_S_MAX_CONCURR_TRX(100) with '...' truncation marker;
  - TIMESTAMP fields show fractions of seconds.

* SYS_VTQ internal InnoDB table;
* I_S.INNODB_SYS_VTQ table;
* vers_notify_vtq(): add record to SYS_VTQ on versioned DML;
* SYS_VTQ columns filled: TRX_ID, BEGIN_TS.

Problem was that we got an error in sequence_insert while opening the
newly created sequence table in an prepared statement as the table id
didn't match.
Fixed by temporarly removing the reprepare observer during
sequence_insert as there can never be a table missmatch in this case.

storage/rocksdb/rdb_datadic.cc failed to compile on big endian machines (wrong usage of static_assert)