main.max_statement_time test not stable on loaded machines, upstream fixes should be backported

[laurynas-biveinis]

For example, an ASan build (which is slower but the test is not skipped, unlike with Valgrind) may fail with:

main.max_statement_time                  w4 [ fail ]
        Test ended at 2015-08-19 23:52:08

CURRENT_TEST: main.max_statement_time
mysqltest: At line 498: query 'SELECT 1 AS STATUS FROM INFORMATION_SCHEMA.GLOBAL_STATUS
WHERE VARIABLE_NAME = 'Qcache_hits'
        AND CONVERT(VARIABLE_VALUE, UNSIGNED) > @qc_hits' failed: 1886: Query execution was interrupted, max_statement_time exceeded

This particular failure could be avoided by resetting max_statement_time for INFORMATION_SCHEMA queries, which shouldn't be subject to timeouts here.

This has been fixed in upstream at 4a4f248. It also has at least one other stability fix (before the feature implementation diverged): a53f8bd. Still the current upstream test case (for the diverged implementation) has at least one instability bug report open: http://bugs.mysql.com/bug.php?id=77959