Memory leaks with UnitOfWork instances

[quantus]

Continuum can leak UnitOfWork instances stored in the VersioningManager. First of all the items are not removed from the dictionary in function VersioningManager.clear. This is simple to fix by adding line del self.units_of_work[conn].

Other issue is that when using continuum with flask-sqlalchemy the function VersioningManager.clear gets called with SignallingSession after the transaction commit phase is completed. VersioningManager uses connections as keys to find the active UnitOfWork instance from unit_of_works dictionary and when SignallingSession's bind value is an Engine instead of Connection the manager fails to find the right unit of work instance. Thus the unit of work instance never gets removed from the dictionary even when the del statement is added as above.

The second issue doesn't happen when the Session is created with a Connection as a bind value, but this isn't possible when using flask-sqlalchemy.

[fake-name]

For the love of god, PLEASE release a new version on PyPi. I've been fighting this issue for months, assuming it was fixed.

Goddammit.

[kvesteri]

Done and sorry for keeping you waiting.

[fake-name]

Awesome. I can remove the git dependency in my build process.

Thanks a bunch!