I often run drop_versioned_table and create_versioned_table to reset a table's versions while in development, as part
of a migration which does other things too.  If other columns are added and removed, sometimes the acts_as_versioned
model can lose track of what columns it actually has before the create_versioned_table method is called, so it winds
up not creating the correct columns in the versioned table.  Attached is a patch which simply calls
self.reset_column_information in create_versioned_table to make sure the column information is back in sync.