Assorted fake changes 5.6 port code review bugs

[Laurynas Biveinis]

Clumping all the issues together to avoid many individual bug reports.

1) row_ins_clust_index_entry(), before row_ins_extern_checkpoint: in case of fake changes, mem_heap_free() should probably be replaced by dtuple_convert_back_big_rec().

2) duplicate ut_ad (... fake changes) in btr_cur_optimistic_insert().

3) row_upd_clust_step(): fake changes handling loses BTR_ALREADY_S_LATCHED from the btr search mode. It is kept in, e.g., row_ins_clust_index_entry_low().

4) row_ins_sec_index_entry_low() should downgrade btr search mode if fake changes enabled.

[Krunal Bauskar (krunal-bauskar)]

#4 was addressed by "bug#1410410: innodb_fake_changes cause unknown error code 1000" ... remaining issue will be address by forthcoming patch meant for this bug.

[Krunal Bauskar (krunal-bauskar)]

commit f516d3c94c2859c6ca4d65213e9f61ffa8108969
Author: Krunal Bauskar <email address hidden>
Date: Wed Sep 16 11:22:36 2015 +0530

    - bug#1204443 : Assorted fake changes 5.6 port code review bugs

      Bug is about spread out changes in fake_changes code path.
      a. correct the memory freeing logic if big-records are involved.
      b. removed duplicate assert validation
      c. update case with online-ddl was not covering fake-change latching logic.

      Added test-cases for missing scenarios.

[Shahriyar Rzayev (rzayev-sehriyar)]

Percona now uses JIRA for bug reports so this bug report is migrated to: https://jira.percona.com/browse/PS-1403