Session/txn retries on sqlalchemy backend

[Joshua Harlow]

When a txn that was started with session.begin() conflicts with another ongoing txn we need to handle this failure (and remerge a new set of data) in the sqlalchemy backend (and repeat this X configurable number of times, then fail if we can not) to further avoid consistency issues.

[Joshua Harlow (harlowja)]

It should be relatively easy to plug-in a retry loop around/in https://github.com/openstack/taskflow/blob/master/taskflow/persistence/backends/impl_sqlalchemy.py#L299 (catching the appropriate exceptions and attempting the functor again...)