Performance: optimize read&&write latency performance

# Description

Read/Write performance bottlenecks often occur in some case:
1. [add](https://github.com/apache/incubator-pegasus/blob/master/scripts/pegasus_add_node_list.sh) or [offline](https://github.com/apache/incubator-pegasus/blob/master/scripts/pegasus_offline_node_list.sh) node
2. long-tail latency in daily case

Here record some sub-issue and pull-request:

- [x] https://github.com/apache/incubator-pegasus/pull/528
- [x] https://github.com/apache/incubator-pegasus/pull/543
- [x] https://github.com/XiaoMi/rdsn/pull/633
- [x] https://github.com/XiaoMi/rdsn/pull/568
- [x] https://github.com/XiaoMi/rdsn/pull/461
- [x] https://github.com/apache/incubator-pegasus/issues/796
https://github.com/XiaoMi/rdsn/pull/909
- [ ] https://github.com/apache/incubator-pegasus/issues/802
- [x] https://github.com/XiaoMi/rdsn/pull/944

Start by reviewing the node-change scripts scripts/pegasus_add_node_list.sh and scripts/pegasus_offline_node_list.sh, then read the linked pull requests and issues, especially the unchecked issue #802. The completed links record prior work on read/write and long-tail latency, but this issue does not define a remaining change or a clear completion condition.

Written by the indexing model from the issue text.