RECOVERY LOSE DATA

Problem:

When too-large blob fields are used, InnoDB overwrites its most recent
checkpoint in its redo logs.  

Solution:

While writing large blobs, call log_free_check() periodically to ensure that
enough free space is there in the redo logs. 

During an update operation, when a non-updated blob column is moved off page
then the update vector is correctly updated before writing the undo log.
This way special handling for update operation is avoided.

rb#4862 approved by Marko.  Yasufumi also contributed (rb#5175).