This is the prerequisite patch to move the sp_instr class and classes derived
from it into the files sp_instr.cc/sp_instr.h. The classes sp_lex_cursor and
sp_lex_keeper have been also moved to the files files sp_instr.cc/sp_instr.h.

Additionally,
  * all occurrences of macroses NULL, FALSE, TRUE have been replaced
    with the corresponding C++ keywords nullptr, false, true.
  * the keyword 'override' has been added in and the keyword 'virtual' removed
    from signatures of every virtual method implemented in classes derived
    from the base class sp_instr.
  * trivial destructors like
     virtual ~sp_instr_stmt() {}
    have been removed for every class derived from the base class sp_instr
    since they are meaningless.
  * added the keyword 'final' into declaration of the class sp_lex_keeper
    since this class shouldn't have a derived class by design.
  * made the function cmp_rqp_locations as static since it is not called
    outside the file sp_instr.cc.
  * moved the function subst_spvars() into the file sp_instr.cc since this
    function is used only by the class sp_instr_stmt that is located
    in the file sp_instr.cc.

This is the prerequisite patch to make interface of the class
Reprepare_observer more like the one used by MySQL.

This patch adds the method can_retry() to the class Reprepare_observer
that returns true in case max. number of attempts to re-run a failing
statement is not yet reached. To control a number of re-run attempts
already done the data member m_attempt has been introduced. Doing this way,
we encapsulate activity with incrementing a counter on every statement
run and checking whether it reaches a limit or not inside implementation
of the class Reprepare_observer instead duplicating such boiler plate code
in every place where controlling for reaching a limit of max. number attempts
for re-running failed statement is required.

commit_try_rebuild(): Only invoke trx_t::drop_table_statistics()
if both InnoDB statistics tables are accessible (and exclusively
locked by the current transaction). This avoids a crash due to
ut_a(sym_node->table != NULL) in pars_retrieve_table_def().

The crash was repeated on a partial copy of a MariaDB 10.3 data
directory that lacked the *.ibd files for the statistics tables.

fil_space_t::flush<false>(): If the CLOSING flag is set,
the file may already have been closed, resulting in EBADF
being returned by fdatasync(). In any case, the
thread that had set the flag should take care of invoking
os_file_flush_func().

The crash occurred during the execution of FLUSH TABLES...FOR EXPORT.

Tested by: Matthias Leich

Also refer to the service file for startup.

Also include the ftp.mariadb.org script rather than old name.

This turns out to be a too expensive call to use for every freed page,
especially under global buffer pool mutex protection.

Also, the usefulness of proactive shrinking of the buffer pool on Windows
this is seems questionable - there is no OOM killer, and the OS will
shrink working sets itself.

This fixes up 25b91c3f

galera_sr.GCF-1060 'innodb'              w2 [ fail ]  timeout after 900 seconds

galera_3nodes.galera_ssl_reload At line 50: mysql_shutdown failed
galera_3nodes.galera_ssl_reload : MDEV-30172 At line 50: mysql_shutdown failed
galera_3nodes.GCF-354 : mysqltest: At line 39: query 'DROP TABLE test.t1' failed: 1047: WSREP has not yet prepared node for application use
galera_3nodes.GCF-354 : mysqltest: At line 30: query 'INSERT INTO test.t1 values (1)' failed: 1180: Got error 6 "No such device or address"

galera_wan : [ERROR] WSREP: /home/buildbot/buildbot/build/gcs/src/gcs_state_msg.cpp:gcs_state_msg_get_quorum():947: Failed to establish quorum.