currently, BR restore has two main steps.

1. perform All prepare DDL(create table + alter auto_increment).
2. restore data

For now, step.2 will wait for step.1 finished. so the time cost will be (step.1 + step.2), unfortunately
perform per DDL will cost 1~2s in TiDB, and we can't do parallel perform due to TiDB restriction.
but we pipeline above steps to reduce cost to max(step.1, step.2).

Following things to do:

• BR should retry RegionError on BatchSplitRegions  #219 Add retry on batchRegionSplit.
• pipeline two steps.
• finish test report.