With this small change, we will almost instantly filter classes and tear down to the Sequel level. This allow us to only 
 `select()` Sequel models instead of the whole Rails workspace. Thus, we do not need to cache the variable. Because, Rails uses eager load or may use threaded model. This means, when this method fired, only few models would be loaded at that time. This will lead missing models for tables.

Note that we also used `c.implicit_table_name`. This is necesary for who uses (like us) multiple Sequel Databases at the same time on single Rails App. `DBModel::TableModel.klazz` will require `implicit_table_name` to demodulize, undersrcore and proper pluralization.

Sequel can send `drop` on commit by default. This can be activated via `on_commit: :drop`. This will minimize queries sent to the Server.

We do not need `self.` calls here. They are redundant.

We also added an exception handling here. Problem here is that, when trigger executed and temporary table does not exists (for example we forgot to send :bulk_audit commands) then there is a issue raises. In this case, if anything goes wrong and transaction is not found, whole object will return immediately without breaking anything.

It seems your tests didn't liked this...