Fix memory issues with large number of tables

[julien-duponchelle]

This the pull request is about memory leak reported by @YvesDup in #57.

I think this need to be merge in binlogstream before merge of the pull request.

[julien-duponchelle]

I have backported the changes to binlogstream. For me it's ok with Python 2.7 and Python 3.3 and with MySQL 5.7.

Someone else?

Thanks a lot @YvesDup for this great patch

[bjoernhaeuser]

We should add some more tests to that case, right?

[YvesDup]

In the new version of BinLogStreamReader class, the "__manage_if_last_event_of_statement" method is never call. This method has to be used to manage items of the two dictionaries, table_map and dict_table_id.

My favorite solution is to implement a "yield" in replacement of "return" in the "fetchone" method. Because after the yield you can call the "__manage_if_last_event_of_statement" method as you can see in YieldBinLogStreamReader class.
An another solution is possible but is more difficult to implement (add more self variables).

And I am wondering how to get back the last safe position of binlog as to restart without error. Any idea ?
May be is not the good place to talk about that ?

About new tests, I did not see event about "alter" order on table (in the examples of project).

[julien-duponchelle]

Too old I 'm not going to rebase it