MDEV-4725: Incorrect binlog state recovery if crash while writing event group

The binlog state was not recovered correctly if XA is not used (eg. InnoDB
disabled), or if server crashed in the middle of writing an event group to the
binlog.

With this patch, we ensure that recovery of binlog state is done even if we do
not do the full XA binlog recovery, and we ensure that we only recover fully
written event groups into the binlog state.

- Make log_slow_statement() always call delete_explain_query().

There were some places where insufficient locking between
parallel threads could cause invalid memory accesses and
possibly other grief.

This patch adds the missing locking, and moves the locking
into the struct rpl_binlog_state methods to make it easier
to see that proper locking is in place everywhere.

Make sure to clear the temporary error before the retry.

Allow only one level of symlink recursion in mysql_tzdata_to_sql,
to avoid infinite loops.

Apply fix suggested by Igor:
- When eliminate_item_equal() generates pair-wise equalities from a 
  multi-equality,  do generate a "bridge" equality between the first 
  field inside SJM nest and the field that's first in the overall multi-equality.

MDEV-5113 Wrong result (extra row) and valgrind warnings in Item_maxmin_subselect::any_value on 2nd execution of PS with SELECT subquery

When setting Item_func_not_all::test_sum_item or Item_func_not_all::test_sub_item,
reset the other one to NULL - they can never be set both. When a PS is reexecuted,
different executions might be optimized differently and a wrong test_su*_item
might stay set from the previous execution.

- MIN/MAX optimizer does a check whether a "field CMP const" comparison uses a constant 
  that's longer than the field it is compared to.  Make this check only for string columns,
  also compare character lengths, not byte lengths.

- Address input from the mail list: change how EXPLAIN is 
  formatted in the slow query log.

The used_tables attribute must be recalculated for the HAVING condition
if the condition is applied to the rows read from a temporary table.

allocate thd_proc_info string in thd memroot, not on the stack, so that it
won't be overwritten while another thread might be printing it

sort Com_ counters. No simple fix for Binlog_ variables.

MDEV-5153: Server crashes in Item_ref::fix_fields on 2nd execution of PS with LEFT JOIN and MERGE view or SELECT SQ

1. Transformation of row IN subquery made the same as single value.
2. replace_where_subcondition() made working on several layers of OR/AND because it called on expression before fix_fields().

(and valgrind warnings)

* move thd userstat initialization to the same function
  that was adding thd userstat to global counters.
* initialize thd->start_bytes_received in THD::init
  (when thd->userstat_running is set)

Singular Item_equal support added.

The problem was that during constant table substitution Item_equal become containing only one constant which was not supported internally.

mysql_upgrade should do --skip-write-binlog by default

by convention query execution time should be 0 if its start_time is 0
(this was lost when fixing MDEV-4578)

port mysqld_safe numa extensions from percona-server: --flush-caches and --numa-interleave

1. use an appropriate errno code
2. put a comma between the errno and the error message text

before running mysql_install_db check for the existence
of $datadir/mysql, not simply $datadir ($datadir might be mounted on
a separate device - exists, but empty)

fix tests where a path was used as a regex.
revert changes to sys_vars.character_sets_dir_basic - we don't run it on windows anyway

quote the path when using it in a regex - the path might contain wildcards (e.g. +)

expand fullhostname inside the string, to have an empty string, not nothing,
when fullhostname is not defined

Modified according to the review comment