DDL drop database/table

[siddontang]

Now the DDL flow for drop database/table is public -> write only -> delete only -> delete all data -> absent.

If the data is very large, we may wait much time for this DDL. If the drop DDL enters delete only state, we can think all servers can't operate this database/table, and the newer database/table with same name will use another ID, so it is safe to enter absent state after delete only state.

And we will send a delete data task to a task list, a leader task runner get this task and do the delete work asynchronously in the backend.

We should save this task in a list like DDL job, so if the task runner down, another runner can re-handle this task again. (we can use DDL owner to handle this task directly.)

TODO:

• implement a task list like DDL job list in meta
• after delete only, save a delete task in task list, finish DDL.
• DDL owner get the delete task and run it, remove it when finished.
• show statistics can show current some delete task status.

@qiuyesuifeng , any to add?