Use CI to continuously monitor the scheduling efficiency in typical scenarios.

Now we only have some tests for scheduling correctness, but no test case or CI for its efficiency. We should consider constructing some tests with reference to the following scenarios:

1. add/remove a tikv-server node.
2. restart all tikv-servers (at almost the same time).
3. rolling restart all tikv-servers.
4. tikv-server reports busy.
5. hot regions split fast.
6. increase/decrease replica count.
...